xref: /openbmc/linux/drivers/infiniband/hw/mlx4/main.c (revision 84d517f3)
1 /*
2  * Copyright (c) 2006, 2007 Cisco Systems, Inc. All rights reserved.
3  * Copyright (c) 2007, 2008 Mellanox Technologies. All rights reserved.
4  *
5  * This software is available to you under a choice of one of two
6  * licenses.  You may choose to be licensed under the terms of the GNU
7  * General Public License (GPL) Version 2, available from the file
8  * COPYING in the main directory of this source tree, or the
9  * OpenIB.org BSD license below:
10  *
11  *     Redistribution and use in source and binary forms, with or
12  *     without modification, are permitted provided that the following
13  *     conditions are met:
14  *
15  *      - Redistributions of source code must retain the above
16  *        copyright notice, this list of conditions and the following
17  *        disclaimer.
18  *
19  *      - Redistributions in binary form must reproduce the above
20  *        copyright notice, this list of conditions and the following
21  *        disclaimer in the documentation and/or other materials
22  *        provided with the distribution.
23  *
24  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
25  * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
26  * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
27  * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
28  * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
29  * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
30  * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
31  * SOFTWARE.
32  */
33 
34 #include <linux/module.h>
35 #include <linux/init.h>
36 #include <linux/slab.h>
37 #include <linux/errno.h>
38 #include <linux/netdevice.h>
39 #include <linux/inetdevice.h>
40 #include <linux/rtnetlink.h>
41 #include <linux/if_vlan.h>
42 #include <net/ipv6.h>
43 #include <net/addrconf.h>
44 
45 #include <rdma/ib_smi.h>
46 #include <rdma/ib_user_verbs.h>
47 #include <rdma/ib_addr.h>
48 
49 #include <linux/mlx4/driver.h>
50 #include <linux/mlx4/cmd.h>
51 #include <linux/mlx4/qp.h>
52 
53 #include "mlx4_ib.h"
54 #include "user.h"
55 
56 #define DRV_NAME	MLX4_IB_DRV_NAME
57 #define DRV_VERSION	"2.2-1"
58 #define DRV_RELDATE	"Feb 2014"
59 
60 #define MLX4_IB_FLOW_MAX_PRIO 0xFFF
61 #define MLX4_IB_FLOW_QPN_MASK 0xFFFFFF
62 
63 MODULE_AUTHOR("Roland Dreier");
64 MODULE_DESCRIPTION("Mellanox ConnectX HCA InfiniBand driver");
65 MODULE_LICENSE("Dual BSD/GPL");
66 MODULE_VERSION(DRV_VERSION);
67 
68 int mlx4_ib_sm_guid_assign = 1;
69 module_param_named(sm_guid_assign, mlx4_ib_sm_guid_assign, int, 0444);
70 MODULE_PARM_DESC(sm_guid_assign, "Enable SM alias_GUID assignment if sm_guid_assign > 0 (Default: 1)");
71 
72 static const char mlx4_ib_version[] =
73 	DRV_NAME ": Mellanox ConnectX InfiniBand driver v"
74 	DRV_VERSION " (" DRV_RELDATE ")\n";
75 
76 struct update_gid_work {
77 	struct work_struct	work;
78 	union ib_gid		gids[128];
79 	struct mlx4_ib_dev     *dev;
80 	int			port;
81 };
82 
83 static void do_slave_init(struct mlx4_ib_dev *ibdev, int slave, int do_init);
84 
85 static struct workqueue_struct *wq;
86 
87 static void init_query_mad(struct ib_smp *mad)
88 {
89 	mad->base_version  = 1;
90 	mad->mgmt_class    = IB_MGMT_CLASS_SUBN_LID_ROUTED;
91 	mad->class_version = 1;
92 	mad->method	   = IB_MGMT_METHOD_GET;
93 }
94 
95 static union ib_gid zgid;
96 
97 static int check_flow_steering_support(struct mlx4_dev *dev)
98 {
99 	int eth_num_ports = 0;
100 	int ib_num_ports = 0;
101 
102 	int dmfs = dev->caps.steering_mode == MLX4_STEERING_MODE_DEVICE_MANAGED;
103 
104 	if (dmfs) {
105 		int i;
106 		mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_ETH)
107 			eth_num_ports++;
108 		mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_IB)
109 			ib_num_ports++;
110 		dmfs &= (!ib_num_ports ||
111 			 (dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_DMFS_IPOIB)) &&
112 			(!eth_num_ports ||
113 			 (dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_FS_EN));
114 		if (ib_num_ports && mlx4_is_mfunc(dev)) {
115 			pr_warn("Device managed flow steering is unavailable for IB port in multifunction env.\n");
116 			dmfs = 0;
117 		}
118 	}
119 	return dmfs;
120 }
121 
122 static int mlx4_ib_query_device(struct ib_device *ibdev,
123 				struct ib_device_attr *props)
124 {
125 	struct mlx4_ib_dev *dev = to_mdev(ibdev);
126 	struct ib_smp *in_mad  = NULL;
127 	struct ib_smp *out_mad = NULL;
128 	int err = -ENOMEM;
129 
130 	in_mad  = kzalloc(sizeof *in_mad, GFP_KERNEL);
131 	out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
132 	if (!in_mad || !out_mad)
133 		goto out;
134 
135 	init_query_mad(in_mad);
136 	in_mad->attr_id = IB_SMP_ATTR_NODE_INFO;
137 
138 	err = mlx4_MAD_IFC(to_mdev(ibdev), MLX4_MAD_IFC_IGNORE_KEYS,
139 			   1, NULL, NULL, in_mad, out_mad);
140 	if (err)
141 		goto out;
142 
143 	memset(props, 0, sizeof *props);
144 
145 	props->fw_ver = dev->dev->caps.fw_ver;
146 	props->device_cap_flags    = IB_DEVICE_CHANGE_PHY_PORT |
147 		IB_DEVICE_PORT_ACTIVE_EVENT		|
148 		IB_DEVICE_SYS_IMAGE_GUID		|
149 		IB_DEVICE_RC_RNR_NAK_GEN		|
150 		IB_DEVICE_BLOCK_MULTICAST_LOOPBACK;
151 	if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_BAD_PKEY_CNTR)
152 		props->device_cap_flags |= IB_DEVICE_BAD_PKEY_CNTR;
153 	if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_BAD_QKEY_CNTR)
154 		props->device_cap_flags |= IB_DEVICE_BAD_QKEY_CNTR;
155 	if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_APM)
156 		props->device_cap_flags |= IB_DEVICE_AUTO_PATH_MIG;
157 	if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_UD_AV_PORT)
158 		props->device_cap_flags |= IB_DEVICE_UD_AV_PORT_ENFORCE;
159 	if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_IPOIB_CSUM)
160 		props->device_cap_flags |= IB_DEVICE_UD_IP_CSUM;
161 	if (dev->dev->caps.max_gso_sz && dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_BLH)
162 		props->device_cap_flags |= IB_DEVICE_UD_TSO;
163 	if (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_RESERVED_LKEY)
164 		props->device_cap_flags |= IB_DEVICE_LOCAL_DMA_LKEY;
165 	if ((dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_LOCAL_INV) &&
166 	    (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_REMOTE_INV) &&
167 	    (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_FAST_REG_WR))
168 		props->device_cap_flags |= IB_DEVICE_MEM_MGT_EXTENSIONS;
169 	if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_XRC)
170 		props->device_cap_flags |= IB_DEVICE_XRC;
171 	if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_MEM_WINDOW)
172 		props->device_cap_flags |= IB_DEVICE_MEM_WINDOW;
173 	if (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_TYPE_2_WIN) {
174 		if (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_WIN_TYPE_2B)
175 			props->device_cap_flags |= IB_DEVICE_MEM_WINDOW_TYPE_2B;
176 		else
177 			props->device_cap_flags |= IB_DEVICE_MEM_WINDOW_TYPE_2A;
178 	if (dev->steering_support ==  MLX4_STEERING_MODE_DEVICE_MANAGED)
179 		props->device_cap_flags |= IB_DEVICE_MANAGED_FLOW_STEERING;
180 	}
181 
182 	props->vendor_id	   = be32_to_cpup((__be32 *) (out_mad->data + 36)) &
183 		0xffffff;
184 	props->vendor_part_id	   = dev->dev->pdev->device;
185 	props->hw_ver		   = be32_to_cpup((__be32 *) (out_mad->data + 32));
186 	memcpy(&props->sys_image_guid, out_mad->data +	4, 8);
187 
188 	props->max_mr_size	   = ~0ull;
189 	props->page_size_cap	   = dev->dev->caps.page_size_cap;
190 	props->max_qp		   = dev->dev->quotas.qp;
191 	props->max_qp_wr	   = dev->dev->caps.max_wqes - MLX4_IB_SQ_MAX_SPARE;
192 	props->max_sge		   = min(dev->dev->caps.max_sq_sg,
193 					 dev->dev->caps.max_rq_sg);
194 	props->max_cq		   = dev->dev->quotas.cq;
195 	props->max_cqe		   = dev->dev->caps.max_cqes;
196 	props->max_mr		   = dev->dev->quotas.mpt;
197 	props->max_pd		   = dev->dev->caps.num_pds - dev->dev->caps.reserved_pds;
198 	props->max_qp_rd_atom	   = dev->dev->caps.max_qp_dest_rdma;
199 	props->max_qp_init_rd_atom = dev->dev->caps.max_qp_init_rdma;
200 	props->max_res_rd_atom	   = props->max_qp_rd_atom * props->max_qp;
201 	props->max_srq		   = dev->dev->quotas.srq;
202 	props->max_srq_wr	   = dev->dev->caps.max_srq_wqes - 1;
203 	props->max_srq_sge	   = dev->dev->caps.max_srq_sge;
204 	props->max_fast_reg_page_list_len = MLX4_MAX_FAST_REG_PAGES;
205 	props->local_ca_ack_delay  = dev->dev->caps.local_ca_ack_delay;
206 	props->atomic_cap	   = dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_ATOMIC ?
207 		IB_ATOMIC_HCA : IB_ATOMIC_NONE;
208 	props->masked_atomic_cap   = props->atomic_cap;
209 	props->max_pkeys	   = dev->dev->caps.pkey_table_len[1];
210 	props->max_mcast_grp	   = dev->dev->caps.num_mgms + dev->dev->caps.num_amgms;
211 	props->max_mcast_qp_attach = dev->dev->caps.num_qp_per_mgm;
212 	props->max_total_mcast_qp_attach = props->max_mcast_qp_attach *
213 					   props->max_mcast_grp;
214 	props->max_map_per_fmr = dev->dev->caps.max_fmr_maps;
215 
216 out:
217 	kfree(in_mad);
218 	kfree(out_mad);
219 
220 	return err;
221 }
222 
223 static enum rdma_link_layer
224 mlx4_ib_port_link_layer(struct ib_device *device, u8 port_num)
225 {
226 	struct mlx4_dev *dev = to_mdev(device)->dev;
227 
228 	return dev->caps.port_mask[port_num] == MLX4_PORT_TYPE_IB ?
229 		IB_LINK_LAYER_INFINIBAND : IB_LINK_LAYER_ETHERNET;
230 }
231 
232 static int ib_link_query_port(struct ib_device *ibdev, u8 port,
233 			      struct ib_port_attr *props, int netw_view)
234 {
235 	struct ib_smp *in_mad  = NULL;
236 	struct ib_smp *out_mad = NULL;
237 	int ext_active_speed;
238 	int mad_ifc_flags = MLX4_MAD_IFC_IGNORE_KEYS;
239 	int err = -ENOMEM;
240 
241 	in_mad  = kzalloc(sizeof *in_mad, GFP_KERNEL);
242 	out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
243 	if (!in_mad || !out_mad)
244 		goto out;
245 
246 	init_query_mad(in_mad);
247 	in_mad->attr_id  = IB_SMP_ATTR_PORT_INFO;
248 	in_mad->attr_mod = cpu_to_be32(port);
249 
250 	if (mlx4_is_mfunc(to_mdev(ibdev)->dev) && netw_view)
251 		mad_ifc_flags |= MLX4_MAD_IFC_NET_VIEW;
252 
253 	err = mlx4_MAD_IFC(to_mdev(ibdev), mad_ifc_flags, port, NULL, NULL,
254 				in_mad, out_mad);
255 	if (err)
256 		goto out;
257 
258 
259 	props->lid		= be16_to_cpup((__be16 *) (out_mad->data + 16));
260 	props->lmc		= out_mad->data[34] & 0x7;
261 	props->sm_lid		= be16_to_cpup((__be16 *) (out_mad->data + 18));
262 	props->sm_sl		= out_mad->data[36] & 0xf;
263 	props->state		= out_mad->data[32] & 0xf;
264 	props->phys_state	= out_mad->data[33] >> 4;
265 	props->port_cap_flags	= be32_to_cpup((__be32 *) (out_mad->data + 20));
266 	if (netw_view)
267 		props->gid_tbl_len = out_mad->data[50];
268 	else
269 		props->gid_tbl_len = to_mdev(ibdev)->dev->caps.gid_table_len[port];
270 	props->max_msg_sz	= to_mdev(ibdev)->dev->caps.max_msg_sz;
271 	props->pkey_tbl_len	= to_mdev(ibdev)->dev->caps.pkey_table_len[port];
272 	props->bad_pkey_cntr	= be16_to_cpup((__be16 *) (out_mad->data + 46));
273 	props->qkey_viol_cntr	= be16_to_cpup((__be16 *) (out_mad->data + 48));
274 	props->active_width	= out_mad->data[31] & 0xf;
275 	props->active_speed	= out_mad->data[35] >> 4;
276 	props->max_mtu		= out_mad->data[41] & 0xf;
277 	props->active_mtu	= out_mad->data[36] >> 4;
278 	props->subnet_timeout	= out_mad->data[51] & 0x1f;
279 	props->max_vl_num	= out_mad->data[37] >> 4;
280 	props->init_type_reply	= out_mad->data[41] >> 4;
281 
282 	/* Check if extended speeds (EDR/FDR/...) are supported */
283 	if (props->port_cap_flags & IB_PORT_EXTENDED_SPEEDS_SUP) {
284 		ext_active_speed = out_mad->data[62] >> 4;
285 
286 		switch (ext_active_speed) {
287 		case 1:
288 			props->active_speed = IB_SPEED_FDR;
289 			break;
290 		case 2:
291 			props->active_speed = IB_SPEED_EDR;
292 			break;
293 		}
294 	}
295 
296 	/* If reported active speed is QDR, check if is FDR-10 */
297 	if (props->active_speed == IB_SPEED_QDR) {
298 		init_query_mad(in_mad);
299 		in_mad->attr_id = MLX4_ATTR_EXTENDED_PORT_INFO;
300 		in_mad->attr_mod = cpu_to_be32(port);
301 
302 		err = mlx4_MAD_IFC(to_mdev(ibdev), mad_ifc_flags, port,
303 				   NULL, NULL, in_mad, out_mad);
304 		if (err)
305 			goto out;
306 
307 		/* Checking LinkSpeedActive for FDR-10 */
308 		if (out_mad->data[15] & 0x1)
309 			props->active_speed = IB_SPEED_FDR10;
310 	}
311 
312 	/* Avoid wrong speed value returned by FW if the IB link is down. */
313 	if (props->state == IB_PORT_DOWN)
314 		 props->active_speed = IB_SPEED_SDR;
315 
316 out:
317 	kfree(in_mad);
318 	kfree(out_mad);
319 	return err;
320 }
321 
322 static u8 state_to_phys_state(enum ib_port_state state)
323 {
324 	return state == IB_PORT_ACTIVE ? 5 : 3;
325 }
326 
327 static int eth_link_query_port(struct ib_device *ibdev, u8 port,
328 			       struct ib_port_attr *props, int netw_view)
329 {
330 
331 	struct mlx4_ib_dev *mdev = to_mdev(ibdev);
332 	struct mlx4_ib_iboe *iboe = &mdev->iboe;
333 	struct net_device *ndev;
334 	enum ib_mtu tmp;
335 	struct mlx4_cmd_mailbox *mailbox;
336 	int err = 0;
337 
338 	mailbox = mlx4_alloc_cmd_mailbox(mdev->dev);
339 	if (IS_ERR(mailbox))
340 		return PTR_ERR(mailbox);
341 
342 	err = mlx4_cmd_box(mdev->dev, 0, mailbox->dma, port, 0,
343 			   MLX4_CMD_QUERY_PORT, MLX4_CMD_TIME_CLASS_B,
344 			   MLX4_CMD_WRAPPED);
345 	if (err)
346 		goto out;
347 
348 	props->active_width	=  (((u8 *)mailbox->buf)[5] == 0x40) ?
349 						IB_WIDTH_4X : IB_WIDTH_1X;
350 	props->active_speed	= IB_SPEED_QDR;
351 	props->port_cap_flags	= IB_PORT_CM_SUP | IB_PORT_IP_BASED_GIDS;
352 	props->gid_tbl_len	= mdev->dev->caps.gid_table_len[port];
353 	props->max_msg_sz	= mdev->dev->caps.max_msg_sz;
354 	props->pkey_tbl_len	= 1;
355 	props->max_mtu		= IB_MTU_4096;
356 	props->max_vl_num	= 2;
357 	props->state		= IB_PORT_DOWN;
358 	props->phys_state	= state_to_phys_state(props->state);
359 	props->active_mtu	= IB_MTU_256;
360 	spin_lock(&iboe->lock);
361 	ndev = iboe->netdevs[port - 1];
362 	if (!ndev)
363 		goto out_unlock;
364 
365 	tmp = iboe_get_mtu(ndev->mtu);
366 	props->active_mtu = tmp ? min(props->max_mtu, tmp) : IB_MTU_256;
367 
368 	props->state		= (netif_running(ndev) && netif_carrier_ok(ndev)) ?
369 					IB_PORT_ACTIVE : IB_PORT_DOWN;
370 	props->phys_state	= state_to_phys_state(props->state);
371 out_unlock:
372 	spin_unlock(&iboe->lock);
373 out:
374 	mlx4_free_cmd_mailbox(mdev->dev, mailbox);
375 	return err;
376 }
377 
378 int __mlx4_ib_query_port(struct ib_device *ibdev, u8 port,
379 			 struct ib_port_attr *props, int netw_view)
380 {
381 	int err;
382 
383 	memset(props, 0, sizeof *props);
384 
385 	err = mlx4_ib_port_link_layer(ibdev, port) == IB_LINK_LAYER_INFINIBAND ?
386 		ib_link_query_port(ibdev, port, props, netw_view) :
387 				eth_link_query_port(ibdev, port, props, netw_view);
388 
389 	return err;
390 }
391 
392 static int mlx4_ib_query_port(struct ib_device *ibdev, u8 port,
393 			      struct ib_port_attr *props)
394 {
395 	/* returns host view */
396 	return __mlx4_ib_query_port(ibdev, port, props, 0);
397 }
398 
399 int __mlx4_ib_query_gid(struct ib_device *ibdev, u8 port, int index,
400 			union ib_gid *gid, int netw_view)
401 {
402 	struct ib_smp *in_mad  = NULL;
403 	struct ib_smp *out_mad = NULL;
404 	int err = -ENOMEM;
405 	struct mlx4_ib_dev *dev = to_mdev(ibdev);
406 	int clear = 0;
407 	int mad_ifc_flags = MLX4_MAD_IFC_IGNORE_KEYS;
408 
409 	in_mad  = kzalloc(sizeof *in_mad, GFP_KERNEL);
410 	out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
411 	if (!in_mad || !out_mad)
412 		goto out;
413 
414 	init_query_mad(in_mad);
415 	in_mad->attr_id  = IB_SMP_ATTR_PORT_INFO;
416 	in_mad->attr_mod = cpu_to_be32(port);
417 
418 	if (mlx4_is_mfunc(dev->dev) && netw_view)
419 		mad_ifc_flags |= MLX4_MAD_IFC_NET_VIEW;
420 
421 	err = mlx4_MAD_IFC(dev, mad_ifc_flags, port, NULL, NULL, in_mad, out_mad);
422 	if (err)
423 		goto out;
424 
425 	memcpy(gid->raw, out_mad->data + 8, 8);
426 
427 	if (mlx4_is_mfunc(dev->dev) && !netw_view) {
428 		if (index) {
429 			/* For any index > 0, return the null guid */
430 			err = 0;
431 			clear = 1;
432 			goto out;
433 		}
434 	}
435 
436 	init_query_mad(in_mad);
437 	in_mad->attr_id  = IB_SMP_ATTR_GUID_INFO;
438 	in_mad->attr_mod = cpu_to_be32(index / 8);
439 
440 	err = mlx4_MAD_IFC(dev, mad_ifc_flags, port,
441 			   NULL, NULL, in_mad, out_mad);
442 	if (err)
443 		goto out;
444 
445 	memcpy(gid->raw + 8, out_mad->data + (index % 8) * 8, 8);
446 
447 out:
448 	if (clear)
449 		memset(gid->raw + 8, 0, 8);
450 	kfree(in_mad);
451 	kfree(out_mad);
452 	return err;
453 }
454 
455 static int iboe_query_gid(struct ib_device *ibdev, u8 port, int index,
456 			  union ib_gid *gid)
457 {
458 	struct mlx4_ib_dev *dev = to_mdev(ibdev);
459 
460 	*gid = dev->iboe.gid_table[port - 1][index];
461 
462 	return 0;
463 }
464 
465 static int mlx4_ib_query_gid(struct ib_device *ibdev, u8 port, int index,
466 			     union ib_gid *gid)
467 {
468 	if (rdma_port_get_link_layer(ibdev, port) == IB_LINK_LAYER_INFINIBAND)
469 		return __mlx4_ib_query_gid(ibdev, port, index, gid, 0);
470 	else
471 		return iboe_query_gid(ibdev, port, index, gid);
472 }
473 
474 int __mlx4_ib_query_pkey(struct ib_device *ibdev, u8 port, u16 index,
475 			 u16 *pkey, int netw_view)
476 {
477 	struct ib_smp *in_mad  = NULL;
478 	struct ib_smp *out_mad = NULL;
479 	int mad_ifc_flags = MLX4_MAD_IFC_IGNORE_KEYS;
480 	int err = -ENOMEM;
481 
482 	in_mad  = kzalloc(sizeof *in_mad, GFP_KERNEL);
483 	out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
484 	if (!in_mad || !out_mad)
485 		goto out;
486 
487 	init_query_mad(in_mad);
488 	in_mad->attr_id  = IB_SMP_ATTR_PKEY_TABLE;
489 	in_mad->attr_mod = cpu_to_be32(index / 32);
490 
491 	if (mlx4_is_mfunc(to_mdev(ibdev)->dev) && netw_view)
492 		mad_ifc_flags |= MLX4_MAD_IFC_NET_VIEW;
493 
494 	err = mlx4_MAD_IFC(to_mdev(ibdev), mad_ifc_flags, port, NULL, NULL,
495 			   in_mad, out_mad);
496 	if (err)
497 		goto out;
498 
499 	*pkey = be16_to_cpu(((__be16 *) out_mad->data)[index % 32]);
500 
501 out:
502 	kfree(in_mad);
503 	kfree(out_mad);
504 	return err;
505 }
506 
507 static int mlx4_ib_query_pkey(struct ib_device *ibdev, u8 port, u16 index, u16 *pkey)
508 {
509 	return __mlx4_ib_query_pkey(ibdev, port, index, pkey, 0);
510 }
511 
512 static int mlx4_ib_modify_device(struct ib_device *ibdev, int mask,
513 				 struct ib_device_modify *props)
514 {
515 	struct mlx4_cmd_mailbox *mailbox;
516 	unsigned long flags;
517 
518 	if (mask & ~IB_DEVICE_MODIFY_NODE_DESC)
519 		return -EOPNOTSUPP;
520 
521 	if (!(mask & IB_DEVICE_MODIFY_NODE_DESC))
522 		return 0;
523 
524 	if (mlx4_is_slave(to_mdev(ibdev)->dev))
525 		return -EOPNOTSUPP;
526 
527 	spin_lock_irqsave(&to_mdev(ibdev)->sm_lock, flags);
528 	memcpy(ibdev->node_desc, props->node_desc, 64);
529 	spin_unlock_irqrestore(&to_mdev(ibdev)->sm_lock, flags);
530 
531 	/*
532 	 * If possible, pass node desc to FW, so it can generate
533 	 * a 144 trap.  If cmd fails, just ignore.
534 	 */
535 	mailbox = mlx4_alloc_cmd_mailbox(to_mdev(ibdev)->dev);
536 	if (IS_ERR(mailbox))
537 		return 0;
538 
539 	memcpy(mailbox->buf, props->node_desc, 64);
540 	mlx4_cmd(to_mdev(ibdev)->dev, mailbox->dma, 1, 0,
541 		 MLX4_CMD_SET_NODE, MLX4_CMD_TIME_CLASS_A, MLX4_CMD_NATIVE);
542 
543 	mlx4_free_cmd_mailbox(to_mdev(ibdev)->dev, mailbox);
544 
545 	return 0;
546 }
547 
548 static int mlx4_SET_PORT(struct mlx4_ib_dev *dev, u8 port, int reset_qkey_viols,
549 			 u32 cap_mask)
550 {
551 	struct mlx4_cmd_mailbox *mailbox;
552 	int err;
553 	u8 is_eth = dev->dev->caps.port_type[port] == MLX4_PORT_TYPE_ETH;
554 
555 	mailbox = mlx4_alloc_cmd_mailbox(dev->dev);
556 	if (IS_ERR(mailbox))
557 		return PTR_ERR(mailbox);
558 
559 	if (dev->dev->flags & MLX4_FLAG_OLD_PORT_CMDS) {
560 		*(u8 *) mailbox->buf	     = !!reset_qkey_viols << 6;
561 		((__be32 *) mailbox->buf)[2] = cpu_to_be32(cap_mask);
562 	} else {
563 		((u8 *) mailbox->buf)[3]     = !!reset_qkey_viols;
564 		((__be32 *) mailbox->buf)[1] = cpu_to_be32(cap_mask);
565 	}
566 
567 	err = mlx4_cmd(dev->dev, mailbox->dma, port, is_eth, MLX4_CMD_SET_PORT,
568 		       MLX4_CMD_TIME_CLASS_B, MLX4_CMD_NATIVE);
569 
570 	mlx4_free_cmd_mailbox(dev->dev, mailbox);
571 	return err;
572 }
573 
574 static int mlx4_ib_modify_port(struct ib_device *ibdev, u8 port, int mask,
575 			       struct ib_port_modify *props)
576 {
577 	struct ib_port_attr attr;
578 	u32 cap_mask;
579 	int err;
580 
581 	mutex_lock(&to_mdev(ibdev)->cap_mask_mutex);
582 
583 	err = mlx4_ib_query_port(ibdev, port, &attr);
584 	if (err)
585 		goto out;
586 
587 	cap_mask = (attr.port_cap_flags | props->set_port_cap_mask) &
588 		~props->clr_port_cap_mask;
589 
590 	err = mlx4_SET_PORT(to_mdev(ibdev), port,
591 			    !!(mask & IB_PORT_RESET_QKEY_CNTR),
592 			    cap_mask);
593 
594 out:
595 	mutex_unlock(&to_mdev(ibdev)->cap_mask_mutex);
596 	return err;
597 }
598 
599 static struct ib_ucontext *mlx4_ib_alloc_ucontext(struct ib_device *ibdev,
600 						  struct ib_udata *udata)
601 {
602 	struct mlx4_ib_dev *dev = to_mdev(ibdev);
603 	struct mlx4_ib_ucontext *context;
604 	struct mlx4_ib_alloc_ucontext_resp_v3 resp_v3;
605 	struct mlx4_ib_alloc_ucontext_resp resp;
606 	int err;
607 
608 	if (!dev->ib_active)
609 		return ERR_PTR(-EAGAIN);
610 
611 	if (ibdev->uverbs_abi_ver == MLX4_IB_UVERBS_NO_DEV_CAPS_ABI_VERSION) {
612 		resp_v3.qp_tab_size      = dev->dev->caps.num_qps;
613 		resp_v3.bf_reg_size      = dev->dev->caps.bf_reg_size;
614 		resp_v3.bf_regs_per_page = dev->dev->caps.bf_regs_per_page;
615 	} else {
616 		resp.dev_caps	      = dev->dev->caps.userspace_caps;
617 		resp.qp_tab_size      = dev->dev->caps.num_qps;
618 		resp.bf_reg_size      = dev->dev->caps.bf_reg_size;
619 		resp.bf_regs_per_page = dev->dev->caps.bf_regs_per_page;
620 		resp.cqe_size	      = dev->dev->caps.cqe_size;
621 	}
622 
623 	context = kmalloc(sizeof *context, GFP_KERNEL);
624 	if (!context)
625 		return ERR_PTR(-ENOMEM);
626 
627 	err = mlx4_uar_alloc(to_mdev(ibdev)->dev, &context->uar);
628 	if (err) {
629 		kfree(context);
630 		return ERR_PTR(err);
631 	}
632 
633 	INIT_LIST_HEAD(&context->db_page_list);
634 	mutex_init(&context->db_page_mutex);
635 
636 	if (ibdev->uverbs_abi_ver == MLX4_IB_UVERBS_NO_DEV_CAPS_ABI_VERSION)
637 		err = ib_copy_to_udata(udata, &resp_v3, sizeof(resp_v3));
638 	else
639 		err = ib_copy_to_udata(udata, &resp, sizeof(resp));
640 
641 	if (err) {
642 		mlx4_uar_free(to_mdev(ibdev)->dev, &context->uar);
643 		kfree(context);
644 		return ERR_PTR(-EFAULT);
645 	}
646 
647 	return &context->ibucontext;
648 }
649 
650 static int mlx4_ib_dealloc_ucontext(struct ib_ucontext *ibcontext)
651 {
652 	struct mlx4_ib_ucontext *context = to_mucontext(ibcontext);
653 
654 	mlx4_uar_free(to_mdev(ibcontext->device)->dev, &context->uar);
655 	kfree(context);
656 
657 	return 0;
658 }
659 
660 static int mlx4_ib_mmap(struct ib_ucontext *context, struct vm_area_struct *vma)
661 {
662 	struct mlx4_ib_dev *dev = to_mdev(context->device);
663 
664 	if (vma->vm_end - vma->vm_start != PAGE_SIZE)
665 		return -EINVAL;
666 
667 	if (vma->vm_pgoff == 0) {
668 		vma->vm_page_prot = pgprot_noncached(vma->vm_page_prot);
669 
670 		if (io_remap_pfn_range(vma, vma->vm_start,
671 				       to_mucontext(context)->uar.pfn,
672 				       PAGE_SIZE, vma->vm_page_prot))
673 			return -EAGAIN;
674 	} else if (vma->vm_pgoff == 1 && dev->dev->caps.bf_reg_size != 0) {
675 		vma->vm_page_prot = pgprot_writecombine(vma->vm_page_prot);
676 
677 		if (io_remap_pfn_range(vma, vma->vm_start,
678 				       to_mucontext(context)->uar.pfn +
679 				       dev->dev->caps.num_uars,
680 				       PAGE_SIZE, vma->vm_page_prot))
681 			return -EAGAIN;
682 	} else
683 		return -EINVAL;
684 
685 	return 0;
686 }
687 
688 static struct ib_pd *mlx4_ib_alloc_pd(struct ib_device *ibdev,
689 				      struct ib_ucontext *context,
690 				      struct ib_udata *udata)
691 {
692 	struct mlx4_ib_pd *pd;
693 	int err;
694 
695 	pd = kmalloc(sizeof *pd, GFP_KERNEL);
696 	if (!pd)
697 		return ERR_PTR(-ENOMEM);
698 
699 	err = mlx4_pd_alloc(to_mdev(ibdev)->dev, &pd->pdn);
700 	if (err) {
701 		kfree(pd);
702 		return ERR_PTR(err);
703 	}
704 
705 	if (context)
706 		if (ib_copy_to_udata(udata, &pd->pdn, sizeof (__u32))) {
707 			mlx4_pd_free(to_mdev(ibdev)->dev, pd->pdn);
708 			kfree(pd);
709 			return ERR_PTR(-EFAULT);
710 		}
711 
712 	return &pd->ibpd;
713 }
714 
715 static int mlx4_ib_dealloc_pd(struct ib_pd *pd)
716 {
717 	mlx4_pd_free(to_mdev(pd->device)->dev, to_mpd(pd)->pdn);
718 	kfree(pd);
719 
720 	return 0;
721 }
722 
723 static struct ib_xrcd *mlx4_ib_alloc_xrcd(struct ib_device *ibdev,
724 					  struct ib_ucontext *context,
725 					  struct ib_udata *udata)
726 {
727 	struct mlx4_ib_xrcd *xrcd;
728 	int err;
729 
730 	if (!(to_mdev(ibdev)->dev->caps.flags & MLX4_DEV_CAP_FLAG_XRC))
731 		return ERR_PTR(-ENOSYS);
732 
733 	xrcd = kmalloc(sizeof *xrcd, GFP_KERNEL);
734 	if (!xrcd)
735 		return ERR_PTR(-ENOMEM);
736 
737 	err = mlx4_xrcd_alloc(to_mdev(ibdev)->dev, &xrcd->xrcdn);
738 	if (err)
739 		goto err1;
740 
741 	xrcd->pd = ib_alloc_pd(ibdev);
742 	if (IS_ERR(xrcd->pd)) {
743 		err = PTR_ERR(xrcd->pd);
744 		goto err2;
745 	}
746 
747 	xrcd->cq = ib_create_cq(ibdev, NULL, NULL, xrcd, 1, 0);
748 	if (IS_ERR(xrcd->cq)) {
749 		err = PTR_ERR(xrcd->cq);
750 		goto err3;
751 	}
752 
753 	return &xrcd->ibxrcd;
754 
755 err3:
756 	ib_dealloc_pd(xrcd->pd);
757 err2:
758 	mlx4_xrcd_free(to_mdev(ibdev)->dev, xrcd->xrcdn);
759 err1:
760 	kfree(xrcd);
761 	return ERR_PTR(err);
762 }
763 
764 static int mlx4_ib_dealloc_xrcd(struct ib_xrcd *xrcd)
765 {
766 	ib_destroy_cq(to_mxrcd(xrcd)->cq);
767 	ib_dealloc_pd(to_mxrcd(xrcd)->pd);
768 	mlx4_xrcd_free(to_mdev(xrcd->device)->dev, to_mxrcd(xrcd)->xrcdn);
769 	kfree(xrcd);
770 
771 	return 0;
772 }
773 
774 static int add_gid_entry(struct ib_qp *ibqp, union ib_gid *gid)
775 {
776 	struct mlx4_ib_qp *mqp = to_mqp(ibqp);
777 	struct mlx4_ib_dev *mdev = to_mdev(ibqp->device);
778 	struct mlx4_ib_gid_entry *ge;
779 
780 	ge = kzalloc(sizeof *ge, GFP_KERNEL);
781 	if (!ge)
782 		return -ENOMEM;
783 
784 	ge->gid = *gid;
785 	if (mlx4_ib_add_mc(mdev, mqp, gid)) {
786 		ge->port = mqp->port;
787 		ge->added = 1;
788 	}
789 
790 	mutex_lock(&mqp->mutex);
791 	list_add_tail(&ge->list, &mqp->gid_list);
792 	mutex_unlock(&mqp->mutex);
793 
794 	return 0;
795 }
796 
797 int mlx4_ib_add_mc(struct mlx4_ib_dev *mdev, struct mlx4_ib_qp *mqp,
798 		   union ib_gid *gid)
799 {
800 	struct net_device *ndev;
801 	int ret = 0;
802 
803 	if (!mqp->port)
804 		return 0;
805 
806 	spin_lock(&mdev->iboe.lock);
807 	ndev = mdev->iboe.netdevs[mqp->port - 1];
808 	if (ndev)
809 		dev_hold(ndev);
810 	spin_unlock(&mdev->iboe.lock);
811 
812 	if (ndev) {
813 		ret = 1;
814 		dev_put(ndev);
815 	}
816 
817 	return ret;
818 }
819 
820 struct mlx4_ib_steering {
821 	struct list_head list;
822 	u64 reg_id;
823 	union ib_gid gid;
824 };
825 
826 static int parse_flow_attr(struct mlx4_dev *dev,
827 			   u32 qp_num,
828 			   union ib_flow_spec *ib_spec,
829 			   struct _rule_hw *mlx4_spec)
830 {
831 	enum mlx4_net_trans_rule_id type;
832 
833 	switch (ib_spec->type) {
834 	case IB_FLOW_SPEC_ETH:
835 		type = MLX4_NET_TRANS_RULE_ID_ETH;
836 		memcpy(mlx4_spec->eth.dst_mac, ib_spec->eth.val.dst_mac,
837 		       ETH_ALEN);
838 		memcpy(mlx4_spec->eth.dst_mac_msk, ib_spec->eth.mask.dst_mac,
839 		       ETH_ALEN);
840 		mlx4_spec->eth.vlan_tag = ib_spec->eth.val.vlan_tag;
841 		mlx4_spec->eth.vlan_tag_msk = ib_spec->eth.mask.vlan_tag;
842 		break;
843 	case IB_FLOW_SPEC_IB:
844 		type = MLX4_NET_TRANS_RULE_ID_IB;
845 		mlx4_spec->ib.l3_qpn =
846 			cpu_to_be32(qp_num);
847 		mlx4_spec->ib.qpn_mask =
848 			cpu_to_be32(MLX4_IB_FLOW_QPN_MASK);
849 		break;
850 
851 
852 	case IB_FLOW_SPEC_IPV4:
853 		type = MLX4_NET_TRANS_RULE_ID_IPV4;
854 		mlx4_spec->ipv4.src_ip = ib_spec->ipv4.val.src_ip;
855 		mlx4_spec->ipv4.src_ip_msk = ib_spec->ipv4.mask.src_ip;
856 		mlx4_spec->ipv4.dst_ip = ib_spec->ipv4.val.dst_ip;
857 		mlx4_spec->ipv4.dst_ip_msk = ib_spec->ipv4.mask.dst_ip;
858 		break;
859 
860 	case IB_FLOW_SPEC_TCP:
861 	case IB_FLOW_SPEC_UDP:
862 		type = ib_spec->type == IB_FLOW_SPEC_TCP ?
863 					MLX4_NET_TRANS_RULE_ID_TCP :
864 					MLX4_NET_TRANS_RULE_ID_UDP;
865 		mlx4_spec->tcp_udp.dst_port = ib_spec->tcp_udp.val.dst_port;
866 		mlx4_spec->tcp_udp.dst_port_msk = ib_spec->tcp_udp.mask.dst_port;
867 		mlx4_spec->tcp_udp.src_port = ib_spec->tcp_udp.val.src_port;
868 		mlx4_spec->tcp_udp.src_port_msk = ib_spec->tcp_udp.mask.src_port;
869 		break;
870 
871 	default:
872 		return -EINVAL;
873 	}
874 	if (mlx4_map_sw_to_hw_steering_id(dev, type) < 0 ||
875 	    mlx4_hw_rule_sz(dev, type) < 0)
876 		return -EINVAL;
877 	mlx4_spec->id = cpu_to_be16(mlx4_map_sw_to_hw_steering_id(dev, type));
878 	mlx4_spec->size = mlx4_hw_rule_sz(dev, type) >> 2;
879 	return mlx4_hw_rule_sz(dev, type);
880 }
881 
882 struct default_rules {
883 	__u32 mandatory_fields[IB_FLOW_SPEC_SUPPORT_LAYERS];
884 	__u32 mandatory_not_fields[IB_FLOW_SPEC_SUPPORT_LAYERS];
885 	__u32 rules_create_list[IB_FLOW_SPEC_SUPPORT_LAYERS];
886 	__u8  link_layer;
887 };
888 static const struct default_rules default_table[] = {
889 	{
890 		.mandatory_fields = {IB_FLOW_SPEC_IPV4},
891 		.mandatory_not_fields = {IB_FLOW_SPEC_ETH},
892 		.rules_create_list = {IB_FLOW_SPEC_IB},
893 		.link_layer = IB_LINK_LAYER_INFINIBAND
894 	}
895 };
896 
897 static int __mlx4_ib_default_rules_match(struct ib_qp *qp,
898 					 struct ib_flow_attr *flow_attr)
899 {
900 	int i, j, k;
901 	void *ib_flow;
902 	const struct default_rules *pdefault_rules = default_table;
903 	u8 link_layer = rdma_port_get_link_layer(qp->device, flow_attr->port);
904 
905 	for (i = 0; i < sizeof(default_table)/sizeof(default_table[0]); i++,
906 	     pdefault_rules++) {
907 		__u32 field_types[IB_FLOW_SPEC_SUPPORT_LAYERS];
908 		memset(&field_types, 0, sizeof(field_types));
909 
910 		if (link_layer != pdefault_rules->link_layer)
911 			continue;
912 
913 		ib_flow = flow_attr + 1;
914 		/* we assume the specs are sorted */
915 		for (j = 0, k = 0; k < IB_FLOW_SPEC_SUPPORT_LAYERS &&
916 		     j < flow_attr->num_of_specs; k++) {
917 			union ib_flow_spec *current_flow =
918 				(union ib_flow_spec *)ib_flow;
919 
920 			/* same layer but different type */
921 			if (((current_flow->type & IB_FLOW_SPEC_LAYER_MASK) ==
922 			     (pdefault_rules->mandatory_fields[k] &
923 			      IB_FLOW_SPEC_LAYER_MASK)) &&
924 			    (current_flow->type !=
925 			     pdefault_rules->mandatory_fields[k]))
926 				goto out;
927 
928 			/* same layer, try match next one */
929 			if (current_flow->type ==
930 			    pdefault_rules->mandatory_fields[k]) {
931 				j++;
932 				ib_flow +=
933 					((union ib_flow_spec *)ib_flow)->size;
934 			}
935 		}
936 
937 		ib_flow = flow_attr + 1;
938 		for (j = 0; j < flow_attr->num_of_specs;
939 		     j++, ib_flow += ((union ib_flow_spec *)ib_flow)->size)
940 			for (k = 0; k < IB_FLOW_SPEC_SUPPORT_LAYERS; k++)
941 				/* same layer and same type */
942 				if (((union ib_flow_spec *)ib_flow)->type ==
943 				    pdefault_rules->mandatory_not_fields[k])
944 					goto out;
945 
946 		return i;
947 	}
948 out:
949 	return -1;
950 }
951 
952 static int __mlx4_ib_create_default_rules(
953 		struct mlx4_ib_dev *mdev,
954 		struct ib_qp *qp,
955 		const struct default_rules *pdefault_rules,
956 		struct _rule_hw *mlx4_spec) {
957 	int size = 0;
958 	int i;
959 
960 	for (i = 0; i < sizeof(pdefault_rules->rules_create_list)/
961 			sizeof(pdefault_rules->rules_create_list[0]); i++) {
962 		int ret;
963 		union ib_flow_spec ib_spec;
964 		switch (pdefault_rules->rules_create_list[i]) {
965 		case 0:
966 			/* no rule */
967 			continue;
968 		case IB_FLOW_SPEC_IB:
969 			ib_spec.type = IB_FLOW_SPEC_IB;
970 			ib_spec.size = sizeof(struct ib_flow_spec_ib);
971 
972 			break;
973 		default:
974 			/* invalid rule */
975 			return -EINVAL;
976 		}
977 		/* We must put empty rule, qpn is being ignored */
978 		ret = parse_flow_attr(mdev->dev, 0, &ib_spec,
979 				      mlx4_spec);
980 		if (ret < 0) {
981 			pr_info("invalid parsing\n");
982 			return -EINVAL;
983 		}
984 
985 		mlx4_spec = (void *)mlx4_spec + ret;
986 		size += ret;
987 	}
988 	return size;
989 }
990 
991 static int __mlx4_ib_create_flow(struct ib_qp *qp, struct ib_flow_attr *flow_attr,
992 			  int domain,
993 			  enum mlx4_net_trans_promisc_mode flow_type,
994 			  u64 *reg_id)
995 {
996 	int ret, i;
997 	int size = 0;
998 	void *ib_flow;
999 	struct mlx4_ib_dev *mdev = to_mdev(qp->device);
1000 	struct mlx4_cmd_mailbox *mailbox;
1001 	struct mlx4_net_trans_rule_hw_ctrl *ctrl;
1002 	int default_flow;
1003 
1004 	static const u16 __mlx4_domain[] = {
1005 		[IB_FLOW_DOMAIN_USER] = MLX4_DOMAIN_UVERBS,
1006 		[IB_FLOW_DOMAIN_ETHTOOL] = MLX4_DOMAIN_ETHTOOL,
1007 		[IB_FLOW_DOMAIN_RFS] = MLX4_DOMAIN_RFS,
1008 		[IB_FLOW_DOMAIN_NIC] = MLX4_DOMAIN_NIC,
1009 	};
1010 
1011 	if (flow_attr->priority > MLX4_IB_FLOW_MAX_PRIO) {
1012 		pr_err("Invalid priority value %d\n", flow_attr->priority);
1013 		return -EINVAL;
1014 	}
1015 
1016 	if (domain >= IB_FLOW_DOMAIN_NUM) {
1017 		pr_err("Invalid domain value %d\n", domain);
1018 		return -EINVAL;
1019 	}
1020 
1021 	if (mlx4_map_sw_to_hw_steering_mode(mdev->dev, flow_type) < 0)
1022 		return -EINVAL;
1023 
1024 	mailbox = mlx4_alloc_cmd_mailbox(mdev->dev);
1025 	if (IS_ERR(mailbox))
1026 		return PTR_ERR(mailbox);
1027 	ctrl = mailbox->buf;
1028 
1029 	ctrl->prio = cpu_to_be16(__mlx4_domain[domain] |
1030 				 flow_attr->priority);
1031 	ctrl->type = mlx4_map_sw_to_hw_steering_mode(mdev->dev, flow_type);
1032 	ctrl->port = flow_attr->port;
1033 	ctrl->qpn = cpu_to_be32(qp->qp_num);
1034 
1035 	ib_flow = flow_attr + 1;
1036 	size += sizeof(struct mlx4_net_trans_rule_hw_ctrl);
1037 	/* Add default flows */
1038 	default_flow = __mlx4_ib_default_rules_match(qp, flow_attr);
1039 	if (default_flow >= 0) {
1040 		ret = __mlx4_ib_create_default_rules(
1041 				mdev, qp, default_table + default_flow,
1042 				mailbox->buf + size);
1043 		if (ret < 0) {
1044 			mlx4_free_cmd_mailbox(mdev->dev, mailbox);
1045 			return -EINVAL;
1046 		}
1047 		size += ret;
1048 	}
1049 	for (i = 0; i < flow_attr->num_of_specs; i++) {
1050 		ret = parse_flow_attr(mdev->dev, qp->qp_num, ib_flow,
1051 				      mailbox->buf + size);
1052 		if (ret < 0) {
1053 			mlx4_free_cmd_mailbox(mdev->dev, mailbox);
1054 			return -EINVAL;
1055 		}
1056 		ib_flow += ((union ib_flow_spec *) ib_flow)->size;
1057 		size += ret;
1058 	}
1059 
1060 	ret = mlx4_cmd_imm(mdev->dev, mailbox->dma, reg_id, size >> 2, 0,
1061 			   MLX4_QP_FLOW_STEERING_ATTACH, MLX4_CMD_TIME_CLASS_A,
1062 			   MLX4_CMD_NATIVE);
1063 	if (ret == -ENOMEM)
1064 		pr_err("mcg table is full. Fail to register network rule.\n");
1065 	else if (ret == -ENXIO)
1066 		pr_err("Device managed flow steering is disabled. Fail to register network rule.\n");
1067 	else if (ret)
1068 		pr_err("Invalid argumant. Fail to register network rule.\n");
1069 
1070 	mlx4_free_cmd_mailbox(mdev->dev, mailbox);
1071 	return ret;
1072 }
1073 
1074 static int __mlx4_ib_destroy_flow(struct mlx4_dev *dev, u64 reg_id)
1075 {
1076 	int err;
1077 	err = mlx4_cmd(dev, reg_id, 0, 0,
1078 		       MLX4_QP_FLOW_STEERING_DETACH, MLX4_CMD_TIME_CLASS_A,
1079 		       MLX4_CMD_NATIVE);
1080 	if (err)
1081 		pr_err("Fail to detach network rule. registration id = 0x%llx\n",
1082 		       reg_id);
1083 	return err;
1084 }
1085 
1086 static struct ib_flow *mlx4_ib_create_flow(struct ib_qp *qp,
1087 				    struct ib_flow_attr *flow_attr,
1088 				    int domain)
1089 {
1090 	int err = 0, i = 0;
1091 	struct mlx4_ib_flow *mflow;
1092 	enum mlx4_net_trans_promisc_mode type[2];
1093 
1094 	memset(type, 0, sizeof(type));
1095 
1096 	mflow = kzalloc(sizeof(*mflow), GFP_KERNEL);
1097 	if (!mflow) {
1098 		err = -ENOMEM;
1099 		goto err_free;
1100 	}
1101 
1102 	switch (flow_attr->type) {
1103 	case IB_FLOW_ATTR_NORMAL:
1104 		type[0] = MLX4_FS_REGULAR;
1105 		break;
1106 
1107 	case IB_FLOW_ATTR_ALL_DEFAULT:
1108 		type[0] = MLX4_FS_ALL_DEFAULT;
1109 		break;
1110 
1111 	case IB_FLOW_ATTR_MC_DEFAULT:
1112 		type[0] = MLX4_FS_MC_DEFAULT;
1113 		break;
1114 
1115 	case IB_FLOW_ATTR_SNIFFER:
1116 		type[0] = MLX4_FS_UC_SNIFFER;
1117 		type[1] = MLX4_FS_MC_SNIFFER;
1118 		break;
1119 
1120 	default:
1121 		err = -EINVAL;
1122 		goto err_free;
1123 	}
1124 
1125 	while (i < ARRAY_SIZE(type) && type[i]) {
1126 		err = __mlx4_ib_create_flow(qp, flow_attr, domain, type[i],
1127 					    &mflow->reg_id[i]);
1128 		if (err)
1129 			goto err_free;
1130 		i++;
1131 	}
1132 
1133 	return &mflow->ibflow;
1134 
1135 err_free:
1136 	kfree(mflow);
1137 	return ERR_PTR(err);
1138 }
1139 
1140 static int mlx4_ib_destroy_flow(struct ib_flow *flow_id)
1141 {
1142 	int err, ret = 0;
1143 	int i = 0;
1144 	struct mlx4_ib_dev *mdev = to_mdev(flow_id->qp->device);
1145 	struct mlx4_ib_flow *mflow = to_mflow(flow_id);
1146 
1147 	while (i < ARRAY_SIZE(mflow->reg_id) && mflow->reg_id[i]) {
1148 		err = __mlx4_ib_destroy_flow(mdev->dev, mflow->reg_id[i]);
1149 		if (err)
1150 			ret = err;
1151 		i++;
1152 	}
1153 
1154 	kfree(mflow);
1155 	return ret;
1156 }
1157 
1158 static int mlx4_ib_mcg_attach(struct ib_qp *ibqp, union ib_gid *gid, u16 lid)
1159 {
1160 	int err;
1161 	struct mlx4_ib_dev *mdev = to_mdev(ibqp->device);
1162 	struct mlx4_ib_qp *mqp = to_mqp(ibqp);
1163 	u64 reg_id;
1164 	struct mlx4_ib_steering *ib_steering = NULL;
1165 	enum mlx4_protocol prot = (gid->raw[1] == 0x0e) ?
1166 		MLX4_PROT_IB_IPV4 : MLX4_PROT_IB_IPV6;
1167 
1168 	if (mdev->dev->caps.steering_mode ==
1169 	    MLX4_STEERING_MODE_DEVICE_MANAGED) {
1170 		ib_steering = kmalloc(sizeof(*ib_steering), GFP_KERNEL);
1171 		if (!ib_steering)
1172 			return -ENOMEM;
1173 	}
1174 
1175 	err = mlx4_multicast_attach(mdev->dev, &mqp->mqp, gid->raw, mqp->port,
1176 				    !!(mqp->flags &
1177 				       MLX4_IB_QP_BLOCK_MULTICAST_LOOPBACK),
1178 				    prot, &reg_id);
1179 	if (err)
1180 		goto err_malloc;
1181 
1182 	err = add_gid_entry(ibqp, gid);
1183 	if (err)
1184 		goto err_add;
1185 
1186 	if (ib_steering) {
1187 		memcpy(ib_steering->gid.raw, gid->raw, 16);
1188 		ib_steering->reg_id = reg_id;
1189 		mutex_lock(&mqp->mutex);
1190 		list_add(&ib_steering->list, &mqp->steering_rules);
1191 		mutex_unlock(&mqp->mutex);
1192 	}
1193 	return 0;
1194 
1195 err_add:
1196 	mlx4_multicast_detach(mdev->dev, &mqp->mqp, gid->raw,
1197 			      prot, reg_id);
1198 err_malloc:
1199 	kfree(ib_steering);
1200 
1201 	return err;
1202 }
1203 
1204 static struct mlx4_ib_gid_entry *find_gid_entry(struct mlx4_ib_qp *qp, u8 *raw)
1205 {
1206 	struct mlx4_ib_gid_entry *ge;
1207 	struct mlx4_ib_gid_entry *tmp;
1208 	struct mlx4_ib_gid_entry *ret = NULL;
1209 
1210 	list_for_each_entry_safe(ge, tmp, &qp->gid_list, list) {
1211 		if (!memcmp(raw, ge->gid.raw, 16)) {
1212 			ret = ge;
1213 			break;
1214 		}
1215 	}
1216 
1217 	return ret;
1218 }
1219 
1220 static int mlx4_ib_mcg_detach(struct ib_qp *ibqp, union ib_gid *gid, u16 lid)
1221 {
1222 	int err;
1223 	struct mlx4_ib_dev *mdev = to_mdev(ibqp->device);
1224 	struct mlx4_ib_qp *mqp = to_mqp(ibqp);
1225 	struct net_device *ndev;
1226 	struct mlx4_ib_gid_entry *ge;
1227 	u64 reg_id = 0;
1228 	enum mlx4_protocol prot = (gid->raw[1] == 0x0e) ?
1229 		MLX4_PROT_IB_IPV4 : MLX4_PROT_IB_IPV6;
1230 
1231 	if (mdev->dev->caps.steering_mode ==
1232 	    MLX4_STEERING_MODE_DEVICE_MANAGED) {
1233 		struct mlx4_ib_steering *ib_steering;
1234 
1235 		mutex_lock(&mqp->mutex);
1236 		list_for_each_entry(ib_steering, &mqp->steering_rules, list) {
1237 			if (!memcmp(ib_steering->gid.raw, gid->raw, 16)) {
1238 				list_del(&ib_steering->list);
1239 				break;
1240 			}
1241 		}
1242 		mutex_unlock(&mqp->mutex);
1243 		if (&ib_steering->list == &mqp->steering_rules) {
1244 			pr_err("Couldn't find reg_id for mgid. Steering rule is left attached\n");
1245 			return -EINVAL;
1246 		}
1247 		reg_id = ib_steering->reg_id;
1248 		kfree(ib_steering);
1249 	}
1250 
1251 	err = mlx4_multicast_detach(mdev->dev, &mqp->mqp, gid->raw,
1252 				    prot, reg_id);
1253 	if (err)
1254 		return err;
1255 
1256 	mutex_lock(&mqp->mutex);
1257 	ge = find_gid_entry(mqp, gid->raw);
1258 	if (ge) {
1259 		spin_lock(&mdev->iboe.lock);
1260 		ndev = ge->added ? mdev->iboe.netdevs[ge->port - 1] : NULL;
1261 		if (ndev)
1262 			dev_hold(ndev);
1263 		spin_unlock(&mdev->iboe.lock);
1264 		if (ndev)
1265 			dev_put(ndev);
1266 		list_del(&ge->list);
1267 		kfree(ge);
1268 	} else
1269 		pr_warn("could not find mgid entry\n");
1270 
1271 	mutex_unlock(&mqp->mutex);
1272 
1273 	return 0;
1274 }
1275 
1276 static int init_node_data(struct mlx4_ib_dev *dev)
1277 {
1278 	struct ib_smp *in_mad  = NULL;
1279 	struct ib_smp *out_mad = NULL;
1280 	int mad_ifc_flags = MLX4_MAD_IFC_IGNORE_KEYS;
1281 	int err = -ENOMEM;
1282 
1283 	in_mad  = kzalloc(sizeof *in_mad, GFP_KERNEL);
1284 	out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
1285 	if (!in_mad || !out_mad)
1286 		goto out;
1287 
1288 	init_query_mad(in_mad);
1289 	in_mad->attr_id = IB_SMP_ATTR_NODE_DESC;
1290 	if (mlx4_is_master(dev->dev))
1291 		mad_ifc_flags |= MLX4_MAD_IFC_NET_VIEW;
1292 
1293 	err = mlx4_MAD_IFC(dev, mad_ifc_flags, 1, NULL, NULL, in_mad, out_mad);
1294 	if (err)
1295 		goto out;
1296 
1297 	memcpy(dev->ib_dev.node_desc, out_mad->data, 64);
1298 
1299 	in_mad->attr_id = IB_SMP_ATTR_NODE_INFO;
1300 
1301 	err = mlx4_MAD_IFC(dev, mad_ifc_flags, 1, NULL, NULL, in_mad, out_mad);
1302 	if (err)
1303 		goto out;
1304 
1305 	dev->dev->rev_id = be32_to_cpup((__be32 *) (out_mad->data + 32));
1306 	memcpy(&dev->ib_dev.node_guid, out_mad->data + 12, 8);
1307 
1308 out:
1309 	kfree(in_mad);
1310 	kfree(out_mad);
1311 	return err;
1312 }
1313 
1314 static ssize_t show_hca(struct device *device, struct device_attribute *attr,
1315 			char *buf)
1316 {
1317 	struct mlx4_ib_dev *dev =
1318 		container_of(device, struct mlx4_ib_dev, ib_dev.dev);
1319 	return sprintf(buf, "MT%d\n", dev->dev->pdev->device);
1320 }
1321 
1322 static ssize_t show_fw_ver(struct device *device, struct device_attribute *attr,
1323 			   char *buf)
1324 {
1325 	struct mlx4_ib_dev *dev =
1326 		container_of(device, struct mlx4_ib_dev, ib_dev.dev);
1327 	return sprintf(buf, "%d.%d.%d\n", (int) (dev->dev->caps.fw_ver >> 32),
1328 		       (int) (dev->dev->caps.fw_ver >> 16) & 0xffff,
1329 		       (int) dev->dev->caps.fw_ver & 0xffff);
1330 }
1331 
1332 static ssize_t show_rev(struct device *device, struct device_attribute *attr,
1333 			char *buf)
1334 {
1335 	struct mlx4_ib_dev *dev =
1336 		container_of(device, struct mlx4_ib_dev, ib_dev.dev);
1337 	return sprintf(buf, "%x\n", dev->dev->rev_id);
1338 }
1339 
1340 static ssize_t show_board(struct device *device, struct device_attribute *attr,
1341 			  char *buf)
1342 {
1343 	struct mlx4_ib_dev *dev =
1344 		container_of(device, struct mlx4_ib_dev, ib_dev.dev);
1345 	return sprintf(buf, "%.*s\n", MLX4_BOARD_ID_LEN,
1346 		       dev->dev->board_id);
1347 }
1348 
1349 static DEVICE_ATTR(hw_rev,   S_IRUGO, show_rev,    NULL);
1350 static DEVICE_ATTR(fw_ver,   S_IRUGO, show_fw_ver, NULL);
1351 static DEVICE_ATTR(hca_type, S_IRUGO, show_hca,    NULL);
1352 static DEVICE_ATTR(board_id, S_IRUGO, show_board,  NULL);
1353 
1354 static struct device_attribute *mlx4_class_attributes[] = {
1355 	&dev_attr_hw_rev,
1356 	&dev_attr_fw_ver,
1357 	&dev_attr_hca_type,
1358 	&dev_attr_board_id
1359 };
1360 
1361 static void mlx4_addrconf_ifid_eui48(u8 *eui, u16 vlan_id,
1362 				     struct net_device *dev)
1363 {
1364 	memcpy(eui, dev->dev_addr, 3);
1365 	memcpy(eui + 5, dev->dev_addr + 3, 3);
1366 	if (vlan_id < 0x1000) {
1367 		eui[3] = vlan_id >> 8;
1368 		eui[4] = vlan_id & 0xff;
1369 	} else {
1370 		eui[3] = 0xff;
1371 		eui[4] = 0xfe;
1372 	}
1373 	eui[0] ^= 2;
1374 }
1375 
1376 static void update_gids_task(struct work_struct *work)
1377 {
1378 	struct update_gid_work *gw = container_of(work, struct update_gid_work, work);
1379 	struct mlx4_cmd_mailbox *mailbox;
1380 	union ib_gid *gids;
1381 	int err;
1382 	struct mlx4_dev	*dev = gw->dev->dev;
1383 
1384 	mailbox = mlx4_alloc_cmd_mailbox(dev);
1385 	if (IS_ERR(mailbox)) {
1386 		pr_warn("update gid table failed %ld\n", PTR_ERR(mailbox));
1387 		return;
1388 	}
1389 
1390 	gids = mailbox->buf;
1391 	memcpy(gids, gw->gids, sizeof gw->gids);
1392 
1393 	err = mlx4_cmd(dev, mailbox->dma, MLX4_SET_PORT_GID_TABLE << 8 | gw->port,
1394 		       1, MLX4_CMD_SET_PORT, MLX4_CMD_TIME_CLASS_B,
1395 		       MLX4_CMD_WRAPPED);
1396 	if (err)
1397 		pr_warn("set port command failed\n");
1398 	else
1399 		mlx4_ib_dispatch_event(gw->dev, gw->port, IB_EVENT_GID_CHANGE);
1400 
1401 	mlx4_free_cmd_mailbox(dev, mailbox);
1402 	kfree(gw);
1403 }
1404 
1405 static void reset_gids_task(struct work_struct *work)
1406 {
1407 	struct update_gid_work *gw =
1408 			container_of(work, struct update_gid_work, work);
1409 	struct mlx4_cmd_mailbox *mailbox;
1410 	union ib_gid *gids;
1411 	int err;
1412 	struct mlx4_dev	*dev = gw->dev->dev;
1413 
1414 	mailbox = mlx4_alloc_cmd_mailbox(dev);
1415 	if (IS_ERR(mailbox)) {
1416 		pr_warn("reset gid table failed\n");
1417 		goto free;
1418 	}
1419 
1420 	gids = mailbox->buf;
1421 	memcpy(gids, gw->gids, sizeof(gw->gids));
1422 
1423 	if (mlx4_ib_port_link_layer(&gw->dev->ib_dev, gw->port) ==
1424 				    IB_LINK_LAYER_ETHERNET) {
1425 		err = mlx4_cmd(dev, mailbox->dma,
1426 			       MLX4_SET_PORT_GID_TABLE << 8 | gw->port,
1427 			       1, MLX4_CMD_SET_PORT,
1428 			       MLX4_CMD_TIME_CLASS_B,
1429 			       MLX4_CMD_WRAPPED);
1430 		if (err)
1431 			pr_warn(KERN_WARNING
1432 				"set port %d command failed\n", gw->port);
1433 	}
1434 
1435 	mlx4_free_cmd_mailbox(dev, mailbox);
1436 free:
1437 	kfree(gw);
1438 }
1439 
1440 static int update_gid_table(struct mlx4_ib_dev *dev, int port,
1441 			    union ib_gid *gid, int clear,
1442 			    int default_gid)
1443 {
1444 	struct update_gid_work *work;
1445 	int i;
1446 	int need_update = 0;
1447 	int free = -1;
1448 	int found = -1;
1449 	int max_gids;
1450 
1451 	if (default_gid) {
1452 		free = 0;
1453 	} else {
1454 		max_gids = dev->dev->caps.gid_table_len[port];
1455 		for (i = 1; i < max_gids; ++i) {
1456 			if (!memcmp(&dev->iboe.gid_table[port - 1][i], gid,
1457 				    sizeof(*gid)))
1458 				found = i;
1459 
1460 			if (clear) {
1461 				if (found >= 0) {
1462 					need_update = 1;
1463 					dev->iboe.gid_table[port - 1][found] =
1464 						zgid;
1465 					break;
1466 				}
1467 			} else {
1468 				if (found >= 0)
1469 					break;
1470 
1471 				if (free < 0 &&
1472 				    !memcmp(&dev->iboe.gid_table[port - 1][i],
1473 					    &zgid, sizeof(*gid)))
1474 					free = i;
1475 			}
1476 		}
1477 	}
1478 
1479 	if (found == -1 && !clear && free >= 0) {
1480 		dev->iboe.gid_table[port - 1][free] = *gid;
1481 		need_update = 1;
1482 	}
1483 
1484 	if (!need_update)
1485 		return 0;
1486 
1487 	work = kzalloc(sizeof(*work), GFP_ATOMIC);
1488 	if (!work)
1489 		return -ENOMEM;
1490 
1491 	memcpy(work->gids, dev->iboe.gid_table[port - 1], sizeof(work->gids));
1492 	INIT_WORK(&work->work, update_gids_task);
1493 	work->port = port;
1494 	work->dev = dev;
1495 	queue_work(wq, &work->work);
1496 
1497 	return 0;
1498 }
1499 
1500 static void mlx4_make_default_gid(struct  net_device *dev, union ib_gid *gid)
1501 {
1502 	gid->global.subnet_prefix = cpu_to_be64(0xfe80000000000000LL);
1503 	mlx4_addrconf_ifid_eui48(&gid->raw[8], 0xffff, dev);
1504 }
1505 
1506 
1507 static int reset_gid_table(struct mlx4_ib_dev *dev, u8 port)
1508 {
1509 	struct update_gid_work *work;
1510 
1511 	work = kzalloc(sizeof(*work), GFP_ATOMIC);
1512 	if (!work)
1513 		return -ENOMEM;
1514 
1515 	memset(dev->iboe.gid_table[port - 1], 0, sizeof(work->gids));
1516 	memset(work->gids, 0, sizeof(work->gids));
1517 	INIT_WORK(&work->work, reset_gids_task);
1518 	work->dev = dev;
1519 	work->port = port;
1520 	queue_work(wq, &work->work);
1521 	return 0;
1522 }
1523 
1524 static int mlx4_ib_addr_event(int event, struct net_device *event_netdev,
1525 			      struct mlx4_ib_dev *ibdev, union ib_gid *gid)
1526 {
1527 	struct mlx4_ib_iboe *iboe;
1528 	int port = 0;
1529 	struct net_device *real_dev = rdma_vlan_dev_real_dev(event_netdev) ?
1530 				rdma_vlan_dev_real_dev(event_netdev) :
1531 				event_netdev;
1532 	union ib_gid default_gid;
1533 
1534 	mlx4_make_default_gid(real_dev, &default_gid);
1535 
1536 	if (!memcmp(gid, &default_gid, sizeof(*gid)))
1537 		return 0;
1538 
1539 	if (event != NETDEV_DOWN && event != NETDEV_UP)
1540 		return 0;
1541 
1542 	if ((real_dev != event_netdev) &&
1543 	    (event == NETDEV_DOWN) &&
1544 	    rdma_link_local_addr((struct in6_addr *)gid))
1545 		return 0;
1546 
1547 	iboe = &ibdev->iboe;
1548 	spin_lock(&iboe->lock);
1549 
1550 	for (port = 1; port <= ibdev->dev->caps.num_ports; ++port)
1551 		if ((netif_is_bond_master(real_dev) &&
1552 		     (real_dev == iboe->masters[port - 1])) ||
1553 		     (!netif_is_bond_master(real_dev) &&
1554 		     (real_dev == iboe->netdevs[port - 1])))
1555 			update_gid_table(ibdev, port, gid,
1556 					 event == NETDEV_DOWN, 0);
1557 
1558 	spin_unlock(&iboe->lock);
1559 	return 0;
1560 
1561 }
1562 
1563 static u8 mlx4_ib_get_dev_port(struct net_device *dev,
1564 			       struct mlx4_ib_dev *ibdev)
1565 {
1566 	u8 port = 0;
1567 	struct mlx4_ib_iboe *iboe;
1568 	struct net_device *real_dev = rdma_vlan_dev_real_dev(dev) ?
1569 				rdma_vlan_dev_real_dev(dev) : dev;
1570 
1571 	iboe = &ibdev->iboe;
1572 
1573 	for (port = 1; port <= ibdev->dev->caps.num_ports; ++port)
1574 		if ((netif_is_bond_master(real_dev) &&
1575 		     (real_dev == iboe->masters[port - 1])) ||
1576 		     (!netif_is_bond_master(real_dev) &&
1577 		     (real_dev == iboe->netdevs[port - 1])))
1578 			break;
1579 
1580 	if ((port == 0) || (port > ibdev->dev->caps.num_ports))
1581 		return 0;
1582 	else
1583 		return port;
1584 }
1585 
1586 static int mlx4_ib_inet_event(struct notifier_block *this, unsigned long event,
1587 				void *ptr)
1588 {
1589 	struct mlx4_ib_dev *ibdev;
1590 	struct in_ifaddr *ifa = ptr;
1591 	union ib_gid gid;
1592 	struct net_device *event_netdev = ifa->ifa_dev->dev;
1593 
1594 	ipv6_addr_set_v4mapped(ifa->ifa_address, (struct in6_addr *)&gid);
1595 
1596 	ibdev = container_of(this, struct mlx4_ib_dev, iboe.nb_inet);
1597 
1598 	mlx4_ib_addr_event(event, event_netdev, ibdev, &gid);
1599 	return NOTIFY_DONE;
1600 }
1601 
1602 #if IS_ENABLED(CONFIG_IPV6)
1603 static int mlx4_ib_inet6_event(struct notifier_block *this, unsigned long event,
1604 				void *ptr)
1605 {
1606 	struct mlx4_ib_dev *ibdev;
1607 	struct inet6_ifaddr *ifa = ptr;
1608 	union  ib_gid *gid = (union ib_gid *)&ifa->addr;
1609 	struct net_device *event_netdev = ifa->idev->dev;
1610 
1611 	ibdev = container_of(this, struct mlx4_ib_dev, iboe.nb_inet6);
1612 
1613 	mlx4_ib_addr_event(event, event_netdev, ibdev, gid);
1614 	return NOTIFY_DONE;
1615 }
1616 #endif
1617 
1618 #define MLX4_IB_INVALID_MAC	((u64)-1)
1619 static void mlx4_ib_update_qps(struct mlx4_ib_dev *ibdev,
1620 			       struct net_device *dev,
1621 			       int port)
1622 {
1623 	u64 new_smac = 0;
1624 	u64 release_mac = MLX4_IB_INVALID_MAC;
1625 	struct mlx4_ib_qp *qp;
1626 
1627 	read_lock(&dev_base_lock);
1628 	new_smac = mlx4_mac_to_u64(dev->dev_addr);
1629 	read_unlock(&dev_base_lock);
1630 
1631 	mutex_lock(&ibdev->qp1_proxy_lock[port - 1]);
1632 	qp = ibdev->qp1_proxy[port - 1];
1633 	if (qp) {
1634 		int new_smac_index;
1635 		u64 old_smac = qp->pri.smac;
1636 		struct mlx4_update_qp_params update_params;
1637 
1638 		if (new_smac == old_smac)
1639 			goto unlock;
1640 
1641 		new_smac_index = mlx4_register_mac(ibdev->dev, port, new_smac);
1642 
1643 		if (new_smac_index < 0)
1644 			goto unlock;
1645 
1646 		update_params.smac_index = new_smac_index;
1647 		if (mlx4_update_qp(ibdev->dev, &qp->mqp, MLX4_UPDATE_QP_SMAC,
1648 				   &update_params)) {
1649 			release_mac = new_smac;
1650 			goto unlock;
1651 		}
1652 
1653 		qp->pri.smac = new_smac;
1654 		qp->pri.smac_index = new_smac_index;
1655 
1656 		release_mac = old_smac;
1657 	}
1658 
1659 unlock:
1660 	mutex_unlock(&ibdev->qp1_proxy_lock[port - 1]);
1661 	if (release_mac != MLX4_IB_INVALID_MAC)
1662 		mlx4_unregister_mac(ibdev->dev, port, release_mac);
1663 }
1664 
1665 static void mlx4_ib_get_dev_addr(struct net_device *dev,
1666 				 struct mlx4_ib_dev *ibdev, u8 port)
1667 {
1668 	struct in_device *in_dev;
1669 #if IS_ENABLED(CONFIG_IPV6)
1670 	struct inet6_dev *in6_dev;
1671 	union ib_gid  *pgid;
1672 	struct inet6_ifaddr *ifp;
1673 #endif
1674 	union ib_gid gid;
1675 
1676 
1677 	if ((port == 0) || (port > ibdev->dev->caps.num_ports))
1678 		return;
1679 
1680 	/* IPv4 gids */
1681 	in_dev = in_dev_get(dev);
1682 	if (in_dev) {
1683 		for_ifa(in_dev) {
1684 			/*ifa->ifa_address;*/
1685 			ipv6_addr_set_v4mapped(ifa->ifa_address,
1686 					       (struct in6_addr *)&gid);
1687 			update_gid_table(ibdev, port, &gid, 0, 0);
1688 		}
1689 		endfor_ifa(in_dev);
1690 		in_dev_put(in_dev);
1691 	}
1692 #if IS_ENABLED(CONFIG_IPV6)
1693 	/* IPv6 gids */
1694 	in6_dev = in6_dev_get(dev);
1695 	if (in6_dev) {
1696 		read_lock_bh(&in6_dev->lock);
1697 		list_for_each_entry(ifp, &in6_dev->addr_list, if_list) {
1698 			pgid = (union ib_gid *)&ifp->addr;
1699 			update_gid_table(ibdev, port, pgid, 0, 0);
1700 		}
1701 		read_unlock_bh(&in6_dev->lock);
1702 		in6_dev_put(in6_dev);
1703 	}
1704 #endif
1705 }
1706 
1707 static void mlx4_ib_set_default_gid(struct mlx4_ib_dev *ibdev,
1708 				 struct  net_device *dev, u8 port)
1709 {
1710 	union ib_gid gid;
1711 	mlx4_make_default_gid(dev, &gid);
1712 	update_gid_table(ibdev, port, &gid, 0, 1);
1713 }
1714 
1715 static int mlx4_ib_init_gid_table(struct mlx4_ib_dev *ibdev)
1716 {
1717 	struct	net_device *dev;
1718 	struct mlx4_ib_iboe *iboe = &ibdev->iboe;
1719 	int i;
1720 
1721 	for (i = 1; i <= ibdev->num_ports; ++i)
1722 		if (reset_gid_table(ibdev, i))
1723 			return -1;
1724 
1725 	read_lock(&dev_base_lock);
1726 	spin_lock(&iboe->lock);
1727 
1728 	for_each_netdev(&init_net, dev) {
1729 		u8 port = mlx4_ib_get_dev_port(dev, ibdev);
1730 		if (port)
1731 			mlx4_ib_get_dev_addr(dev, ibdev, port);
1732 	}
1733 
1734 	spin_unlock(&iboe->lock);
1735 	read_unlock(&dev_base_lock);
1736 
1737 	return 0;
1738 }
1739 
1740 static void mlx4_ib_scan_netdevs(struct mlx4_ib_dev *ibdev,
1741 				 struct net_device *dev,
1742 				 unsigned long event)
1743 
1744 {
1745 	struct mlx4_ib_iboe *iboe;
1746 	int update_qps_port = -1;
1747 	int port;
1748 
1749 	iboe = &ibdev->iboe;
1750 
1751 	spin_lock(&iboe->lock);
1752 	mlx4_foreach_ib_transport_port(port, ibdev->dev) {
1753 		enum ib_port_state	port_state = IB_PORT_NOP;
1754 		struct net_device *old_master = iboe->masters[port - 1];
1755 		struct net_device *curr_netdev;
1756 		struct net_device *curr_master;
1757 
1758 		iboe->netdevs[port - 1] =
1759 			mlx4_get_protocol_dev(ibdev->dev, MLX4_PROT_ETH, port);
1760 		if (iboe->netdevs[port - 1])
1761 			mlx4_ib_set_default_gid(ibdev,
1762 						iboe->netdevs[port - 1], port);
1763 		curr_netdev = iboe->netdevs[port - 1];
1764 
1765 		if (iboe->netdevs[port - 1] &&
1766 		    netif_is_bond_slave(iboe->netdevs[port - 1])) {
1767 			iboe->masters[port - 1] = netdev_master_upper_dev_get(
1768 				iboe->netdevs[port - 1]);
1769 		} else {
1770 			iboe->masters[port - 1] = NULL;
1771 		}
1772 		curr_master = iboe->masters[port - 1];
1773 
1774 		if (dev == iboe->netdevs[port - 1] &&
1775 		    (event == NETDEV_CHANGEADDR || event == NETDEV_REGISTER ||
1776 		     event == NETDEV_UP || event == NETDEV_CHANGE))
1777 			update_qps_port = port;
1778 
1779 		if (curr_netdev) {
1780 			port_state = (netif_running(curr_netdev) && netif_carrier_ok(curr_netdev)) ?
1781 						IB_PORT_ACTIVE : IB_PORT_DOWN;
1782 			mlx4_ib_set_default_gid(ibdev, curr_netdev, port);
1783 		} else {
1784 			reset_gid_table(ibdev, port);
1785 		}
1786 		/* if using bonding/team and a slave port is down, we don't the bond IP
1787 		 * based gids in the table since flows that select port by gid may get
1788 		 * the down port.
1789 		 */
1790 		if (curr_master && (port_state == IB_PORT_DOWN)) {
1791 			reset_gid_table(ibdev, port);
1792 			mlx4_ib_set_default_gid(ibdev, curr_netdev, port);
1793 		}
1794 		/* if bonding is used it is possible that we add it to masters
1795 		 * only after IP address is assigned to the net bonding
1796 		 * interface.
1797 		*/
1798 		if (curr_master && (old_master != curr_master)) {
1799 			reset_gid_table(ibdev, port);
1800 			mlx4_ib_set_default_gid(ibdev, curr_netdev, port);
1801 			mlx4_ib_get_dev_addr(curr_master, ibdev, port);
1802 		}
1803 
1804 		if (!curr_master && (old_master != curr_master)) {
1805 			reset_gid_table(ibdev, port);
1806 			mlx4_ib_set_default_gid(ibdev, curr_netdev, port);
1807 			mlx4_ib_get_dev_addr(curr_netdev, ibdev, port);
1808 		}
1809 	}
1810 
1811 	spin_unlock(&iboe->lock);
1812 
1813 	if (update_qps_port > 0)
1814 		mlx4_ib_update_qps(ibdev, dev, update_qps_port);
1815 }
1816 
1817 static int mlx4_ib_netdev_event(struct notifier_block *this,
1818 				unsigned long event, void *ptr)
1819 {
1820 	struct net_device *dev = netdev_notifier_info_to_dev(ptr);
1821 	struct mlx4_ib_dev *ibdev;
1822 
1823 	if (!net_eq(dev_net(dev), &init_net))
1824 		return NOTIFY_DONE;
1825 
1826 	ibdev = container_of(this, struct mlx4_ib_dev, iboe.nb);
1827 	mlx4_ib_scan_netdevs(ibdev, dev, event);
1828 
1829 	return NOTIFY_DONE;
1830 }
1831 
1832 static void init_pkeys(struct mlx4_ib_dev *ibdev)
1833 {
1834 	int port;
1835 	int slave;
1836 	int i;
1837 
1838 	if (mlx4_is_master(ibdev->dev)) {
1839 		for (slave = 0; slave <= ibdev->dev->num_vfs; ++slave) {
1840 			for (port = 1; port <= ibdev->dev->caps.num_ports; ++port) {
1841 				for (i = 0;
1842 				     i < ibdev->dev->phys_caps.pkey_phys_table_len[port];
1843 				     ++i) {
1844 					ibdev->pkeys.virt2phys_pkey[slave][port - 1][i] =
1845 					/* master has the identity virt2phys pkey mapping */
1846 						(slave == mlx4_master_func_num(ibdev->dev) || !i) ? i :
1847 							ibdev->dev->phys_caps.pkey_phys_table_len[port] - 1;
1848 					mlx4_sync_pkey_table(ibdev->dev, slave, port, i,
1849 							     ibdev->pkeys.virt2phys_pkey[slave][port - 1][i]);
1850 				}
1851 			}
1852 		}
1853 		/* initialize pkey cache */
1854 		for (port = 1; port <= ibdev->dev->caps.num_ports; ++port) {
1855 			for (i = 0;
1856 			     i < ibdev->dev->phys_caps.pkey_phys_table_len[port];
1857 			     ++i)
1858 				ibdev->pkeys.phys_pkey_cache[port-1][i] =
1859 					(i) ? 0 : 0xFFFF;
1860 		}
1861 	}
1862 }
1863 
1864 static void mlx4_ib_alloc_eqs(struct mlx4_dev *dev, struct mlx4_ib_dev *ibdev)
1865 {
1866 	char name[80];
1867 	int eq_per_port = 0;
1868 	int added_eqs = 0;
1869 	int total_eqs = 0;
1870 	int i, j, eq;
1871 
1872 	/* Legacy mode or comp_pool is not large enough */
1873 	if (dev->caps.comp_pool == 0 ||
1874 	    dev->caps.num_ports > dev->caps.comp_pool)
1875 		return;
1876 
1877 	eq_per_port = rounddown_pow_of_two(dev->caps.comp_pool/
1878 					dev->caps.num_ports);
1879 
1880 	/* Init eq table */
1881 	added_eqs = 0;
1882 	mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_IB)
1883 		added_eqs += eq_per_port;
1884 
1885 	total_eqs = dev->caps.num_comp_vectors + added_eqs;
1886 
1887 	ibdev->eq_table = kzalloc(total_eqs * sizeof(int), GFP_KERNEL);
1888 	if (!ibdev->eq_table)
1889 		return;
1890 
1891 	ibdev->eq_added = added_eqs;
1892 
1893 	eq = 0;
1894 	mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_IB) {
1895 		for (j = 0; j < eq_per_port; j++) {
1896 			snprintf(name, sizeof(name), "mlx4-ib-%d-%d@%s",
1897 				 i, j, dev->pdev->bus->name);
1898 			/* Set IRQ for specific name (per ring) */
1899 			if (mlx4_assign_eq(dev, name, NULL,
1900 					   &ibdev->eq_table[eq])) {
1901 				/* Use legacy (same as mlx4_en driver) */
1902 				pr_warn("Can't allocate EQ %d; reverting to legacy\n", eq);
1903 				ibdev->eq_table[eq] =
1904 					(eq % dev->caps.num_comp_vectors);
1905 			}
1906 			eq++;
1907 		}
1908 	}
1909 
1910 	/* Fill the reset of the vector with legacy EQ */
1911 	for (i = 0, eq = added_eqs; i < dev->caps.num_comp_vectors; i++)
1912 		ibdev->eq_table[eq++] = i;
1913 
1914 	/* Advertise the new number of EQs to clients */
1915 	ibdev->ib_dev.num_comp_vectors = total_eqs;
1916 }
1917 
1918 static void mlx4_ib_free_eqs(struct mlx4_dev *dev, struct mlx4_ib_dev *ibdev)
1919 {
1920 	int i;
1921 
1922 	/* no additional eqs were added */
1923 	if (!ibdev->eq_table)
1924 		return;
1925 
1926 	/* Reset the advertised EQ number */
1927 	ibdev->ib_dev.num_comp_vectors = dev->caps.num_comp_vectors;
1928 
1929 	/* Free only the added eqs */
1930 	for (i = 0; i < ibdev->eq_added; i++) {
1931 		/* Don't free legacy eqs if used */
1932 		if (ibdev->eq_table[i] <= dev->caps.num_comp_vectors)
1933 			continue;
1934 		mlx4_release_eq(dev, ibdev->eq_table[i]);
1935 	}
1936 
1937 	kfree(ibdev->eq_table);
1938 }
1939 
1940 static void *mlx4_ib_add(struct mlx4_dev *dev)
1941 {
1942 	struct mlx4_ib_dev *ibdev;
1943 	int num_ports = 0;
1944 	int i, j;
1945 	int err;
1946 	struct mlx4_ib_iboe *iboe;
1947 	int ib_num_ports = 0;
1948 
1949 	pr_info_once("%s", mlx4_ib_version);
1950 
1951 	num_ports = 0;
1952 	mlx4_foreach_ib_transport_port(i, dev)
1953 		num_ports++;
1954 
1955 	/* No point in registering a device with no ports... */
1956 	if (num_ports == 0)
1957 		return NULL;
1958 
1959 	ibdev = (struct mlx4_ib_dev *) ib_alloc_device(sizeof *ibdev);
1960 	if (!ibdev) {
1961 		dev_err(&dev->pdev->dev, "Device struct alloc failed\n");
1962 		return NULL;
1963 	}
1964 
1965 	iboe = &ibdev->iboe;
1966 
1967 	if (mlx4_pd_alloc(dev, &ibdev->priv_pdn))
1968 		goto err_dealloc;
1969 
1970 	if (mlx4_uar_alloc(dev, &ibdev->priv_uar))
1971 		goto err_pd;
1972 
1973 	ibdev->uar_map = ioremap((phys_addr_t) ibdev->priv_uar.pfn << PAGE_SHIFT,
1974 				 PAGE_SIZE);
1975 	if (!ibdev->uar_map)
1976 		goto err_uar;
1977 	MLX4_INIT_DOORBELL_LOCK(&ibdev->uar_lock);
1978 
1979 	ibdev->dev = dev;
1980 
1981 	strlcpy(ibdev->ib_dev.name, "mlx4_%d", IB_DEVICE_NAME_MAX);
1982 	ibdev->ib_dev.owner		= THIS_MODULE;
1983 	ibdev->ib_dev.node_type		= RDMA_NODE_IB_CA;
1984 	ibdev->ib_dev.local_dma_lkey	= dev->caps.reserved_lkey;
1985 	ibdev->num_ports		= num_ports;
1986 	ibdev->ib_dev.phys_port_cnt     = ibdev->num_ports;
1987 	ibdev->ib_dev.num_comp_vectors	= dev->caps.num_comp_vectors;
1988 	ibdev->ib_dev.dma_device	= &dev->pdev->dev;
1989 
1990 	if (dev->caps.userspace_caps)
1991 		ibdev->ib_dev.uverbs_abi_ver = MLX4_IB_UVERBS_ABI_VERSION;
1992 	else
1993 		ibdev->ib_dev.uverbs_abi_ver = MLX4_IB_UVERBS_NO_DEV_CAPS_ABI_VERSION;
1994 
1995 	ibdev->ib_dev.uverbs_cmd_mask	=
1996 		(1ull << IB_USER_VERBS_CMD_GET_CONTEXT)		|
1997 		(1ull << IB_USER_VERBS_CMD_QUERY_DEVICE)	|
1998 		(1ull << IB_USER_VERBS_CMD_QUERY_PORT)		|
1999 		(1ull << IB_USER_VERBS_CMD_ALLOC_PD)		|
2000 		(1ull << IB_USER_VERBS_CMD_DEALLOC_PD)		|
2001 		(1ull << IB_USER_VERBS_CMD_REG_MR)		|
2002 		(1ull << IB_USER_VERBS_CMD_DEREG_MR)		|
2003 		(1ull << IB_USER_VERBS_CMD_CREATE_COMP_CHANNEL)	|
2004 		(1ull << IB_USER_VERBS_CMD_CREATE_CQ)		|
2005 		(1ull << IB_USER_VERBS_CMD_RESIZE_CQ)		|
2006 		(1ull << IB_USER_VERBS_CMD_DESTROY_CQ)		|
2007 		(1ull << IB_USER_VERBS_CMD_CREATE_QP)		|
2008 		(1ull << IB_USER_VERBS_CMD_MODIFY_QP)		|
2009 		(1ull << IB_USER_VERBS_CMD_QUERY_QP)		|
2010 		(1ull << IB_USER_VERBS_CMD_DESTROY_QP)		|
2011 		(1ull << IB_USER_VERBS_CMD_ATTACH_MCAST)	|
2012 		(1ull << IB_USER_VERBS_CMD_DETACH_MCAST)	|
2013 		(1ull << IB_USER_VERBS_CMD_CREATE_SRQ)		|
2014 		(1ull << IB_USER_VERBS_CMD_MODIFY_SRQ)		|
2015 		(1ull << IB_USER_VERBS_CMD_QUERY_SRQ)		|
2016 		(1ull << IB_USER_VERBS_CMD_DESTROY_SRQ)		|
2017 		(1ull << IB_USER_VERBS_CMD_CREATE_XSRQ)		|
2018 		(1ull << IB_USER_VERBS_CMD_OPEN_QP);
2019 
2020 	ibdev->ib_dev.query_device	= mlx4_ib_query_device;
2021 	ibdev->ib_dev.query_port	= mlx4_ib_query_port;
2022 	ibdev->ib_dev.get_link_layer	= mlx4_ib_port_link_layer;
2023 	ibdev->ib_dev.query_gid		= mlx4_ib_query_gid;
2024 	ibdev->ib_dev.query_pkey	= mlx4_ib_query_pkey;
2025 	ibdev->ib_dev.modify_device	= mlx4_ib_modify_device;
2026 	ibdev->ib_dev.modify_port	= mlx4_ib_modify_port;
2027 	ibdev->ib_dev.alloc_ucontext	= mlx4_ib_alloc_ucontext;
2028 	ibdev->ib_dev.dealloc_ucontext	= mlx4_ib_dealloc_ucontext;
2029 	ibdev->ib_dev.mmap		= mlx4_ib_mmap;
2030 	ibdev->ib_dev.alloc_pd		= mlx4_ib_alloc_pd;
2031 	ibdev->ib_dev.dealloc_pd	= mlx4_ib_dealloc_pd;
2032 	ibdev->ib_dev.create_ah		= mlx4_ib_create_ah;
2033 	ibdev->ib_dev.query_ah		= mlx4_ib_query_ah;
2034 	ibdev->ib_dev.destroy_ah	= mlx4_ib_destroy_ah;
2035 	ibdev->ib_dev.create_srq	= mlx4_ib_create_srq;
2036 	ibdev->ib_dev.modify_srq	= mlx4_ib_modify_srq;
2037 	ibdev->ib_dev.query_srq		= mlx4_ib_query_srq;
2038 	ibdev->ib_dev.destroy_srq	= mlx4_ib_destroy_srq;
2039 	ibdev->ib_dev.post_srq_recv	= mlx4_ib_post_srq_recv;
2040 	ibdev->ib_dev.create_qp		= mlx4_ib_create_qp;
2041 	ibdev->ib_dev.modify_qp		= mlx4_ib_modify_qp;
2042 	ibdev->ib_dev.query_qp		= mlx4_ib_query_qp;
2043 	ibdev->ib_dev.destroy_qp	= mlx4_ib_destroy_qp;
2044 	ibdev->ib_dev.post_send		= mlx4_ib_post_send;
2045 	ibdev->ib_dev.post_recv		= mlx4_ib_post_recv;
2046 	ibdev->ib_dev.create_cq		= mlx4_ib_create_cq;
2047 	ibdev->ib_dev.modify_cq		= mlx4_ib_modify_cq;
2048 	ibdev->ib_dev.resize_cq		= mlx4_ib_resize_cq;
2049 	ibdev->ib_dev.destroy_cq	= mlx4_ib_destroy_cq;
2050 	ibdev->ib_dev.poll_cq		= mlx4_ib_poll_cq;
2051 	ibdev->ib_dev.req_notify_cq	= mlx4_ib_arm_cq;
2052 	ibdev->ib_dev.get_dma_mr	= mlx4_ib_get_dma_mr;
2053 	ibdev->ib_dev.reg_user_mr	= mlx4_ib_reg_user_mr;
2054 	ibdev->ib_dev.dereg_mr		= mlx4_ib_dereg_mr;
2055 	ibdev->ib_dev.alloc_fast_reg_mr = mlx4_ib_alloc_fast_reg_mr;
2056 	ibdev->ib_dev.alloc_fast_reg_page_list = mlx4_ib_alloc_fast_reg_page_list;
2057 	ibdev->ib_dev.free_fast_reg_page_list  = mlx4_ib_free_fast_reg_page_list;
2058 	ibdev->ib_dev.attach_mcast	= mlx4_ib_mcg_attach;
2059 	ibdev->ib_dev.detach_mcast	= mlx4_ib_mcg_detach;
2060 	ibdev->ib_dev.process_mad	= mlx4_ib_process_mad;
2061 
2062 	if (!mlx4_is_slave(ibdev->dev)) {
2063 		ibdev->ib_dev.alloc_fmr		= mlx4_ib_fmr_alloc;
2064 		ibdev->ib_dev.map_phys_fmr	= mlx4_ib_map_phys_fmr;
2065 		ibdev->ib_dev.unmap_fmr		= mlx4_ib_unmap_fmr;
2066 		ibdev->ib_dev.dealloc_fmr	= mlx4_ib_fmr_dealloc;
2067 	}
2068 
2069 	if (dev->caps.flags & MLX4_DEV_CAP_FLAG_MEM_WINDOW ||
2070 	    dev->caps.bmme_flags & MLX4_BMME_FLAG_TYPE_2_WIN) {
2071 		ibdev->ib_dev.alloc_mw = mlx4_ib_alloc_mw;
2072 		ibdev->ib_dev.bind_mw = mlx4_ib_bind_mw;
2073 		ibdev->ib_dev.dealloc_mw = mlx4_ib_dealloc_mw;
2074 
2075 		ibdev->ib_dev.uverbs_cmd_mask |=
2076 			(1ull << IB_USER_VERBS_CMD_ALLOC_MW) |
2077 			(1ull << IB_USER_VERBS_CMD_DEALLOC_MW);
2078 	}
2079 
2080 	if (dev->caps.flags & MLX4_DEV_CAP_FLAG_XRC) {
2081 		ibdev->ib_dev.alloc_xrcd = mlx4_ib_alloc_xrcd;
2082 		ibdev->ib_dev.dealloc_xrcd = mlx4_ib_dealloc_xrcd;
2083 		ibdev->ib_dev.uverbs_cmd_mask |=
2084 			(1ull << IB_USER_VERBS_CMD_OPEN_XRCD) |
2085 			(1ull << IB_USER_VERBS_CMD_CLOSE_XRCD);
2086 	}
2087 
2088 	if (check_flow_steering_support(dev)) {
2089 		ibdev->steering_support = MLX4_STEERING_MODE_DEVICE_MANAGED;
2090 		ibdev->ib_dev.create_flow	= mlx4_ib_create_flow;
2091 		ibdev->ib_dev.destroy_flow	= mlx4_ib_destroy_flow;
2092 
2093 		ibdev->ib_dev.uverbs_ex_cmd_mask	|=
2094 			(1ull << IB_USER_VERBS_EX_CMD_CREATE_FLOW) |
2095 			(1ull << IB_USER_VERBS_EX_CMD_DESTROY_FLOW);
2096 	}
2097 
2098 	mlx4_ib_alloc_eqs(dev, ibdev);
2099 
2100 	spin_lock_init(&iboe->lock);
2101 
2102 	if (init_node_data(ibdev))
2103 		goto err_map;
2104 
2105 	for (i = 0; i < ibdev->num_ports; ++i) {
2106 		mutex_init(&ibdev->qp1_proxy_lock[i]);
2107 		if (mlx4_ib_port_link_layer(&ibdev->ib_dev, i + 1) ==
2108 						IB_LINK_LAYER_ETHERNET) {
2109 			err = mlx4_counter_alloc(ibdev->dev, &ibdev->counters[i]);
2110 			if (err)
2111 				ibdev->counters[i] = -1;
2112 		} else {
2113 			ibdev->counters[i] = -1;
2114 		}
2115 	}
2116 
2117 	mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_IB)
2118 		ib_num_ports++;
2119 
2120 	spin_lock_init(&ibdev->sm_lock);
2121 	mutex_init(&ibdev->cap_mask_mutex);
2122 
2123 	if (ibdev->steering_support == MLX4_STEERING_MODE_DEVICE_MANAGED &&
2124 	    ib_num_ports) {
2125 		ibdev->steer_qpn_count = MLX4_IB_UC_MAX_NUM_QPS;
2126 		err = mlx4_qp_reserve_range(dev, ibdev->steer_qpn_count,
2127 					    MLX4_IB_UC_STEER_QPN_ALIGN,
2128 					    &ibdev->steer_qpn_base);
2129 		if (err)
2130 			goto err_counter;
2131 
2132 		ibdev->ib_uc_qpns_bitmap =
2133 			kmalloc(BITS_TO_LONGS(ibdev->steer_qpn_count) *
2134 				sizeof(long),
2135 				GFP_KERNEL);
2136 		if (!ibdev->ib_uc_qpns_bitmap) {
2137 			dev_err(&dev->pdev->dev, "bit map alloc failed\n");
2138 			goto err_steer_qp_release;
2139 		}
2140 
2141 		bitmap_zero(ibdev->ib_uc_qpns_bitmap, ibdev->steer_qpn_count);
2142 
2143 		err = mlx4_FLOW_STEERING_IB_UC_QP_RANGE(
2144 				dev, ibdev->steer_qpn_base,
2145 				ibdev->steer_qpn_base +
2146 				ibdev->steer_qpn_count - 1);
2147 		if (err)
2148 			goto err_steer_free_bitmap;
2149 	}
2150 
2151 	if (ib_register_device(&ibdev->ib_dev, NULL))
2152 		goto err_steer_free_bitmap;
2153 
2154 	if (mlx4_ib_mad_init(ibdev))
2155 		goto err_reg;
2156 
2157 	if (mlx4_ib_init_sriov(ibdev))
2158 		goto err_mad;
2159 
2160 	if (dev->caps.flags & MLX4_DEV_CAP_FLAG_IBOE) {
2161 		if (!iboe->nb.notifier_call) {
2162 			iboe->nb.notifier_call = mlx4_ib_netdev_event;
2163 			err = register_netdevice_notifier(&iboe->nb);
2164 			if (err) {
2165 				iboe->nb.notifier_call = NULL;
2166 				goto err_notif;
2167 			}
2168 		}
2169 		if (!iboe->nb_inet.notifier_call) {
2170 			iboe->nb_inet.notifier_call = mlx4_ib_inet_event;
2171 			err = register_inetaddr_notifier(&iboe->nb_inet);
2172 			if (err) {
2173 				iboe->nb_inet.notifier_call = NULL;
2174 				goto err_notif;
2175 			}
2176 		}
2177 #if IS_ENABLED(CONFIG_IPV6)
2178 		if (!iboe->nb_inet6.notifier_call) {
2179 			iboe->nb_inet6.notifier_call = mlx4_ib_inet6_event;
2180 			err = register_inet6addr_notifier(&iboe->nb_inet6);
2181 			if (err) {
2182 				iboe->nb_inet6.notifier_call = NULL;
2183 				goto err_notif;
2184 			}
2185 		}
2186 #endif
2187 		for (i = 1 ; i <= ibdev->num_ports ; ++i)
2188 			reset_gid_table(ibdev, i);
2189 		rtnl_lock();
2190 		mlx4_ib_scan_netdevs(ibdev, NULL, 0);
2191 		rtnl_unlock();
2192 		mlx4_ib_init_gid_table(ibdev);
2193 	}
2194 
2195 	for (j = 0; j < ARRAY_SIZE(mlx4_class_attributes); ++j) {
2196 		if (device_create_file(&ibdev->ib_dev.dev,
2197 				       mlx4_class_attributes[j]))
2198 			goto err_notif;
2199 	}
2200 
2201 	ibdev->ib_active = true;
2202 
2203 	if (mlx4_is_mfunc(ibdev->dev))
2204 		init_pkeys(ibdev);
2205 
2206 	/* create paravirt contexts for any VFs which are active */
2207 	if (mlx4_is_master(ibdev->dev)) {
2208 		for (j = 0; j < MLX4_MFUNC_MAX; j++) {
2209 			if (j == mlx4_master_func_num(ibdev->dev))
2210 				continue;
2211 			if (mlx4_is_slave_active(ibdev->dev, j))
2212 				do_slave_init(ibdev, j, 1);
2213 		}
2214 	}
2215 	return ibdev;
2216 
2217 err_notif:
2218 	if (ibdev->iboe.nb.notifier_call) {
2219 		if (unregister_netdevice_notifier(&ibdev->iboe.nb))
2220 			pr_warn("failure unregistering notifier\n");
2221 		ibdev->iboe.nb.notifier_call = NULL;
2222 	}
2223 	if (ibdev->iboe.nb_inet.notifier_call) {
2224 		if (unregister_inetaddr_notifier(&ibdev->iboe.nb_inet))
2225 			pr_warn("failure unregistering notifier\n");
2226 		ibdev->iboe.nb_inet.notifier_call = NULL;
2227 	}
2228 #if IS_ENABLED(CONFIG_IPV6)
2229 	if (ibdev->iboe.nb_inet6.notifier_call) {
2230 		if (unregister_inet6addr_notifier(&ibdev->iboe.nb_inet6))
2231 			pr_warn("failure unregistering notifier\n");
2232 		ibdev->iboe.nb_inet6.notifier_call = NULL;
2233 	}
2234 #endif
2235 	flush_workqueue(wq);
2236 
2237 	mlx4_ib_close_sriov(ibdev);
2238 
2239 err_mad:
2240 	mlx4_ib_mad_cleanup(ibdev);
2241 
2242 err_reg:
2243 	ib_unregister_device(&ibdev->ib_dev);
2244 
2245 err_steer_free_bitmap:
2246 	kfree(ibdev->ib_uc_qpns_bitmap);
2247 
2248 err_steer_qp_release:
2249 	if (ibdev->steering_support == MLX4_STEERING_MODE_DEVICE_MANAGED)
2250 		mlx4_qp_release_range(dev, ibdev->steer_qpn_base,
2251 				      ibdev->steer_qpn_count);
2252 err_counter:
2253 	for (; i; --i)
2254 		if (ibdev->counters[i - 1] != -1)
2255 			mlx4_counter_free(ibdev->dev, ibdev->counters[i - 1]);
2256 
2257 err_map:
2258 	iounmap(ibdev->uar_map);
2259 
2260 err_uar:
2261 	mlx4_uar_free(dev, &ibdev->priv_uar);
2262 
2263 err_pd:
2264 	mlx4_pd_free(dev, ibdev->priv_pdn);
2265 
2266 err_dealloc:
2267 	ib_dealloc_device(&ibdev->ib_dev);
2268 
2269 	return NULL;
2270 }
2271 
2272 int mlx4_ib_steer_qp_alloc(struct mlx4_ib_dev *dev, int count, int *qpn)
2273 {
2274 	int offset;
2275 
2276 	WARN_ON(!dev->ib_uc_qpns_bitmap);
2277 
2278 	offset = bitmap_find_free_region(dev->ib_uc_qpns_bitmap,
2279 					 dev->steer_qpn_count,
2280 					 get_count_order(count));
2281 	if (offset < 0)
2282 		return offset;
2283 
2284 	*qpn = dev->steer_qpn_base + offset;
2285 	return 0;
2286 }
2287 
2288 void mlx4_ib_steer_qp_free(struct mlx4_ib_dev *dev, u32 qpn, int count)
2289 {
2290 	if (!qpn ||
2291 	    dev->steering_support != MLX4_STEERING_MODE_DEVICE_MANAGED)
2292 		return;
2293 
2294 	BUG_ON(qpn < dev->steer_qpn_base);
2295 
2296 	bitmap_release_region(dev->ib_uc_qpns_bitmap,
2297 			      qpn - dev->steer_qpn_base,
2298 			      get_count_order(count));
2299 }
2300 
2301 int mlx4_ib_steer_qp_reg(struct mlx4_ib_dev *mdev, struct mlx4_ib_qp *mqp,
2302 			 int is_attach)
2303 {
2304 	int err;
2305 	size_t flow_size;
2306 	struct ib_flow_attr *flow = NULL;
2307 	struct ib_flow_spec_ib *ib_spec;
2308 
2309 	if (is_attach) {
2310 		flow_size = sizeof(struct ib_flow_attr) +
2311 			    sizeof(struct ib_flow_spec_ib);
2312 		flow = kzalloc(flow_size, GFP_KERNEL);
2313 		if (!flow)
2314 			return -ENOMEM;
2315 		flow->port = mqp->port;
2316 		flow->num_of_specs = 1;
2317 		flow->size = flow_size;
2318 		ib_spec = (struct ib_flow_spec_ib *)(flow + 1);
2319 		ib_spec->type = IB_FLOW_SPEC_IB;
2320 		ib_spec->size = sizeof(struct ib_flow_spec_ib);
2321 		/* Add an empty rule for IB L2 */
2322 		memset(&ib_spec->mask, 0, sizeof(ib_spec->mask));
2323 
2324 		err = __mlx4_ib_create_flow(&mqp->ibqp, flow,
2325 					    IB_FLOW_DOMAIN_NIC,
2326 					    MLX4_FS_REGULAR,
2327 					    &mqp->reg_id);
2328 	} else {
2329 		err = __mlx4_ib_destroy_flow(mdev->dev, mqp->reg_id);
2330 	}
2331 	kfree(flow);
2332 	return err;
2333 }
2334 
2335 static void mlx4_ib_remove(struct mlx4_dev *dev, void *ibdev_ptr)
2336 {
2337 	struct mlx4_ib_dev *ibdev = ibdev_ptr;
2338 	int p;
2339 
2340 	mlx4_ib_close_sriov(ibdev);
2341 	mlx4_ib_mad_cleanup(ibdev);
2342 	ib_unregister_device(&ibdev->ib_dev);
2343 	if (ibdev->iboe.nb.notifier_call) {
2344 		if (unregister_netdevice_notifier(&ibdev->iboe.nb))
2345 			pr_warn("failure unregistering notifier\n");
2346 		ibdev->iboe.nb.notifier_call = NULL;
2347 	}
2348 
2349 	if (ibdev->steering_support == MLX4_STEERING_MODE_DEVICE_MANAGED) {
2350 		mlx4_qp_release_range(dev, ibdev->steer_qpn_base,
2351 				      ibdev->steer_qpn_count);
2352 		kfree(ibdev->ib_uc_qpns_bitmap);
2353 	}
2354 
2355 	if (ibdev->iboe.nb_inet.notifier_call) {
2356 		if (unregister_inetaddr_notifier(&ibdev->iboe.nb_inet))
2357 			pr_warn("failure unregistering notifier\n");
2358 		ibdev->iboe.nb_inet.notifier_call = NULL;
2359 	}
2360 #if IS_ENABLED(CONFIG_IPV6)
2361 	if (ibdev->iboe.nb_inet6.notifier_call) {
2362 		if (unregister_inet6addr_notifier(&ibdev->iboe.nb_inet6))
2363 			pr_warn("failure unregistering notifier\n");
2364 		ibdev->iboe.nb_inet6.notifier_call = NULL;
2365 	}
2366 #endif
2367 
2368 	iounmap(ibdev->uar_map);
2369 	for (p = 0; p < ibdev->num_ports; ++p)
2370 		if (ibdev->counters[p] != -1)
2371 			mlx4_counter_free(ibdev->dev, ibdev->counters[p]);
2372 	mlx4_foreach_port(p, dev, MLX4_PORT_TYPE_IB)
2373 		mlx4_CLOSE_PORT(dev, p);
2374 
2375 	mlx4_ib_free_eqs(dev, ibdev);
2376 
2377 	mlx4_uar_free(dev, &ibdev->priv_uar);
2378 	mlx4_pd_free(dev, ibdev->priv_pdn);
2379 	ib_dealloc_device(&ibdev->ib_dev);
2380 }
2381 
2382 static void do_slave_init(struct mlx4_ib_dev *ibdev, int slave, int do_init)
2383 {
2384 	struct mlx4_ib_demux_work **dm = NULL;
2385 	struct mlx4_dev *dev = ibdev->dev;
2386 	int i;
2387 	unsigned long flags;
2388 	struct mlx4_active_ports actv_ports;
2389 	unsigned int ports;
2390 	unsigned int first_port;
2391 
2392 	if (!mlx4_is_master(dev))
2393 		return;
2394 
2395 	actv_ports = mlx4_get_active_ports(dev, slave);
2396 	ports = bitmap_weight(actv_ports.ports, dev->caps.num_ports);
2397 	first_port = find_first_bit(actv_ports.ports, dev->caps.num_ports);
2398 
2399 	dm = kcalloc(ports, sizeof(*dm), GFP_ATOMIC);
2400 	if (!dm) {
2401 		pr_err("failed to allocate memory for tunneling qp update\n");
2402 		goto out;
2403 	}
2404 
2405 	for (i = 0; i < ports; i++) {
2406 		dm[i] = kmalloc(sizeof (struct mlx4_ib_demux_work), GFP_ATOMIC);
2407 		if (!dm[i]) {
2408 			pr_err("failed to allocate memory for tunneling qp update work struct\n");
2409 			for (i = 0; i < dev->caps.num_ports; i++) {
2410 				if (dm[i])
2411 					kfree(dm[i]);
2412 			}
2413 			goto out;
2414 		}
2415 	}
2416 	/* initialize or tear down tunnel QPs for the slave */
2417 	for (i = 0; i < ports; i++) {
2418 		INIT_WORK(&dm[i]->work, mlx4_ib_tunnels_update_work);
2419 		dm[i]->port = first_port + i + 1;
2420 		dm[i]->slave = slave;
2421 		dm[i]->do_init = do_init;
2422 		dm[i]->dev = ibdev;
2423 		spin_lock_irqsave(&ibdev->sriov.going_down_lock, flags);
2424 		if (!ibdev->sriov.is_going_down)
2425 			queue_work(ibdev->sriov.demux[i].ud_wq, &dm[i]->work);
2426 		spin_unlock_irqrestore(&ibdev->sriov.going_down_lock, flags);
2427 	}
2428 out:
2429 	kfree(dm);
2430 	return;
2431 }
2432 
2433 static void mlx4_ib_event(struct mlx4_dev *dev, void *ibdev_ptr,
2434 			  enum mlx4_dev_event event, unsigned long param)
2435 {
2436 	struct ib_event ibev;
2437 	struct mlx4_ib_dev *ibdev = to_mdev((struct ib_device *) ibdev_ptr);
2438 	struct mlx4_eqe *eqe = NULL;
2439 	struct ib_event_work *ew;
2440 	int p = 0;
2441 
2442 	if (event == MLX4_DEV_EVENT_PORT_MGMT_CHANGE)
2443 		eqe = (struct mlx4_eqe *)param;
2444 	else
2445 		p = (int) param;
2446 
2447 	switch (event) {
2448 	case MLX4_DEV_EVENT_PORT_UP:
2449 		if (p > ibdev->num_ports)
2450 			return;
2451 		if (mlx4_is_master(dev) &&
2452 		    rdma_port_get_link_layer(&ibdev->ib_dev, p) ==
2453 			IB_LINK_LAYER_INFINIBAND) {
2454 			mlx4_ib_invalidate_all_guid_record(ibdev, p);
2455 		}
2456 		ibev.event = IB_EVENT_PORT_ACTIVE;
2457 		break;
2458 
2459 	case MLX4_DEV_EVENT_PORT_DOWN:
2460 		if (p > ibdev->num_ports)
2461 			return;
2462 		ibev.event = IB_EVENT_PORT_ERR;
2463 		break;
2464 
2465 	case MLX4_DEV_EVENT_CATASTROPHIC_ERROR:
2466 		ibdev->ib_active = false;
2467 		ibev.event = IB_EVENT_DEVICE_FATAL;
2468 		break;
2469 
2470 	case MLX4_DEV_EVENT_PORT_MGMT_CHANGE:
2471 		ew = kmalloc(sizeof *ew, GFP_ATOMIC);
2472 		if (!ew) {
2473 			pr_err("failed to allocate memory for events work\n");
2474 			break;
2475 		}
2476 
2477 		INIT_WORK(&ew->work, handle_port_mgmt_change_event);
2478 		memcpy(&ew->ib_eqe, eqe, sizeof *eqe);
2479 		ew->ib_dev = ibdev;
2480 		/* need to queue only for port owner, which uses GEN_EQE */
2481 		if (mlx4_is_master(dev))
2482 			queue_work(wq, &ew->work);
2483 		else
2484 			handle_port_mgmt_change_event(&ew->work);
2485 		return;
2486 
2487 	case MLX4_DEV_EVENT_SLAVE_INIT:
2488 		/* here, p is the slave id */
2489 		do_slave_init(ibdev, p, 1);
2490 		return;
2491 
2492 	case MLX4_DEV_EVENT_SLAVE_SHUTDOWN:
2493 		/* here, p is the slave id */
2494 		do_slave_init(ibdev, p, 0);
2495 		return;
2496 
2497 	default:
2498 		return;
2499 	}
2500 
2501 	ibev.device	      = ibdev_ptr;
2502 	ibev.element.port_num = (u8) p;
2503 
2504 	ib_dispatch_event(&ibev);
2505 }
2506 
2507 static struct mlx4_interface mlx4_ib_interface = {
2508 	.add		= mlx4_ib_add,
2509 	.remove		= mlx4_ib_remove,
2510 	.event		= mlx4_ib_event,
2511 	.protocol	= MLX4_PROT_IB_IPV6
2512 };
2513 
2514 static int __init mlx4_ib_init(void)
2515 {
2516 	int err;
2517 
2518 	wq = create_singlethread_workqueue("mlx4_ib");
2519 	if (!wq)
2520 		return -ENOMEM;
2521 
2522 	err = mlx4_ib_mcg_init();
2523 	if (err)
2524 		goto clean_wq;
2525 
2526 	err = mlx4_register_interface(&mlx4_ib_interface);
2527 	if (err)
2528 		goto clean_mcg;
2529 
2530 	return 0;
2531 
2532 clean_mcg:
2533 	mlx4_ib_mcg_destroy();
2534 
2535 clean_wq:
2536 	destroy_workqueue(wq);
2537 	return err;
2538 }
2539 
2540 static void __exit mlx4_ib_cleanup(void)
2541 {
2542 	mlx4_unregister_interface(&mlx4_ib_interface);
2543 	mlx4_ib_mcg_destroy();
2544 	destroy_workqueue(wq);
2545 }
2546 
2547 module_init(mlx4_ib_init);
2548 module_exit(mlx4_ib_cleanup);
2549